temp

added depth_to_xyz experiment

.vscode/settings.json

@@ -157,7 +157,14 @@
 		"sledge_abi_symbols.h": "c",
 		"mutex": "c",
 		"lock.h": "c",
-		"route_latency.h": "c"
+		"route_latency.h": "c",
+		"wasm_globals.h": "c",
+		"*.bak": "c",
+		"sandbox_set_as_initialized.h": "c",
+		"sandbox_set_as_allocated.h": "c",
+		"sandbox_total.h": "c",
+		"system_error": "c",
+		"metrics_server.h": "c"
 	},
 	"files.exclude": {
 		"**/.git": true,

Makefile

@@ -35,7 +35,7 @@ libsledge.clean:
 .PHONY: runtime
 runtime:
 	make -C runtime
-
+	./rsync.sh
 
 .PHONY: runtime.clean
 runtime.clean:

applications/Makefile

@@ -12,7 +12,7 @@ LDFLAGS=-shared -fPIC -Wl,--export-dynamic,--whole-archive -L../libsledge/dist/
 # LDFLAGS=-flto -fvisibility=hidden
 
 # Strips out calls to assert() and disables debuglog
-CFLAGS+=-DNDEBUG
+# CFLAGS+=-DNDEBUG
 
 dist:
 	mkdir -p dist

libsledge/Makefile

@@ -5,7 +5,7 @@ INCLUDES := -Iinclude/
 CFLAGS := -fPIC -O3 -flto -ftls-model=initial-exec
 
 # Strips out calls to assert() and disables debuglog
-CFLAGS+=-DNDEBUG
+# CFLAGS+=-DNDEBUG
 
 # CFI Sanitizer
 # CFLAGS+=-fvisibility=default -fsanitize=cfi

libsledge/include/sledge_abi.h

@@ -28,6 +28,7 @@ struct sledge_abi__wasm_memory {
 	uint64_t capacity; /* Size backed by actual pages */
 	uint64_t max;      /* Soft cap in bytes. Defaults to 4GB */
 	uint8_t *buffer;   /* Backing heap allocation. Different lifetime because realloc might move this */
+    uint64_t id;
 };
 
 /* This structure is the runtime representation of the unique state of a module instance

libsledge/src/memory_instructions.c

@@ -18,6 +18,7 @@ INLINE float
 get_f32(uint32_t offset)
 {
 	assert(sledge_abi__current_wasm_module_instance.memory.buffer != NULL);
+	assert(offset < sledge_abi__current_wasm_module_instance.memory.size);
 
 	return *(float *)&sledge_abi__current_wasm_module_instance.memory.buffer[offset];
 }
@@ -26,6 +27,7 @@ INLINE double
 get_f64(uint32_t offset)
 {
 	assert(sledge_abi__current_wasm_module_instance.memory.buffer != NULL);
+	assert(offset < sledge_abi__current_wasm_module_instance.memory.size);
 
 	return *(double *)&sledge_abi__current_wasm_module_instance.memory.buffer[offset];
 }
@@ -34,6 +36,7 @@ INLINE int8_t
 get_i8(uint32_t offset)
 {
 	assert(sledge_abi__current_wasm_module_instance.memory.buffer != NULL);
+	assert(offset < sledge_abi__current_wasm_module_instance.memory.size);
 
 	return *(int8_t *)&sledge_abi__current_wasm_module_instance.memory.buffer[offset];
 }
@@ -42,6 +45,7 @@ INLINE int16_t
 get_i16(uint32_t offset)
 {
 	assert(sledge_abi__current_wasm_module_instance.memory.buffer != NULL);
+	assert(offset < sledge_abi__current_wasm_module_instance.memory.size);
 
 	return *(int16_t *)&sledge_abi__current_wasm_module_instance.memory.buffer[offset];
 }
@@ -50,6 +54,7 @@ INLINE int32_t
 get_i32(uint32_t offset)
 {
 	assert(sledge_abi__current_wasm_module_instance.memory.buffer != NULL);
+	assert(offset < sledge_abi__current_wasm_module_instance.memory.size);
 
 	return *(int32_t *)&sledge_abi__current_wasm_module_instance.memory.buffer[offset];
 }
@@ -58,6 +63,7 @@ INLINE int64_t
 get_i64(uint32_t offset)
 {
 	assert(sledge_abi__current_wasm_module_instance.memory.buffer != NULL);
+	assert(offset < sledge_abi__current_wasm_module_instance.memory.size);
 
 	return *(int64_t *)&sledge_abi__current_wasm_module_instance.memory.buffer[offset];
 }
@@ -67,6 +73,7 @@ INLINE void
 set_f32(uint32_t offset, float value)
 {
 	assert(sledge_abi__current_wasm_module_instance.memory.buffer != NULL);
+	assert(offset < sledge_abi__current_wasm_module_instance.memory.size);
 
 	*(float *)&sledge_abi__current_wasm_module_instance.memory.buffer[offset] = value;
 }
@@ -75,6 +82,7 @@ INLINE void
 set_f64(uint32_t offset, double value)
 {
 	assert(sledge_abi__current_wasm_module_instance.memory.buffer != NULL);
+	assert(offset < sledge_abi__current_wasm_module_instance.memory.size);
 
 	*(double *)&sledge_abi__current_wasm_module_instance.memory.buffer[offset] = value;
 }
@@ -83,6 +91,7 @@ INLINE void
 set_i8(uint32_t offset, int8_t value)
 {
 	assert(sledge_abi__current_wasm_module_instance.memory.buffer != NULL);
+	assert(offset < sledge_abi__current_wasm_module_instance.memory.size);
 
 	*(int8_t *)&sledge_abi__current_wasm_module_instance.memory.buffer[offset] = value;
 }
@@ -91,6 +100,7 @@ INLINE void
 set_i16(uint32_t offset, int16_t value)
 {
 	assert(sledge_abi__current_wasm_module_instance.memory.buffer != NULL);
+	assert(offset < sledge_abi__current_wasm_module_instance.memory.size);
 
 	*(int16_t *)&sledge_abi__current_wasm_module_instance.memory.buffer[offset] = value;
 }
@@ -99,6 +109,7 @@ INLINE void
 set_i32(uint32_t offset, int32_t value)
 {
 	assert(sledge_abi__current_wasm_module_instance.memory.buffer != NULL);
+	assert(offset < sledge_abi__current_wasm_module_instance.memory.size);
 
 	*(int32_t *)&sledge_abi__current_wasm_module_instance.memory.buffer[offset] = value;
 }
@@ -107,6 +118,7 @@ INLINE void
 set_i64(uint32_t offset, int64_t value)
 {
 	assert(sledge_abi__current_wasm_module_instance.memory.buffer != NULL);
+	assert(offset < sledge_abi__current_wasm_module_instance.memory.size);
 
 	*(int64_t *)&sledge_abi__current_wasm_module_instance.memory.buffer[offset] = value;
 }

rsync.sh

@@ -0,0 +1,27 @@
+#!/bin/bash
+
+# rsync -ru --progress --exclude={'res','out.txt','out.dat','err.dat'} ./tests/* emil@c220g2-011017.wisc.cloudlab.us:/users/emil/sledge-server/tests/
+# rsync -ru --progress --exclude={'res','out.txt','out.dat','err.dat'} ./tests/* emil@c220g2-011016.wisc.cloudlab.us:/users/emil/sledge-client/tests/
+
+# rsync -ru --progress --exclude={'thirdparty','res','err.dat','out*','*.log'} ./tests ./runtime emil@c220g2-011314.wisc.cloudlab.us:/users/emil/sledge-server/
+# rsync -ru --progress --exclude={'res','err.dat','out*','*.log'} ./tests emil@c220g2-011323.wisc.cloudlab.us:/users/emil/sledge-client/
+
+rsync -ru --progress --exclude={'thirdparty','res-*','err.dat','out*','*.log','input*'} ./tests ./runtime emil@128.105.145.72:/users/emil/sledge-server/
+rsync -ru --progress --exclude={'thirdparty','res-*','err.dat','out*','*.log','mt-juan/input-cnn','mt-emil/input-cnn'} ./tests ./runtime emil@128.105.145.71:/users/emil/sledge-client/
+# rsync -ru --progress --exclude={'thirdparty','res-*','err.dat','out*','*.log','mt-juan/input-cnn','mt-emil/input-cnn'} ./tests ./runtime emil@128.105.145.70:/users/emil/sledge-client/
+rsync -ru --progress --exclude={'thirdparty','res-*','err.dat','out*','*.log','input*'} ./tests ./runtime emil@128.105.145.70:/users/emil/sledge-server/
+# rsync -ru --progress --exclude={'thirdparty','res-*','err.dat','out*','*.log'} ./tests ./runtime emil@128.105.145.132:/users/emil/sledge-client/
+
+# If on a network where only 443 is allowed use this (after allowing port forwarding ssh to 443 on the server):
+# rsync -ru -e 'ssh -p 443' --progress --exclude={'res','out.txt','out.dat','err.dat'} ./tests/* emil@server:/users/emil/sledge-server/tests/
+# rsync -ru -e 'ssh -p 443' --progress --exclude={'res','out.txt','out.dat','err.dat'} ./tests/* emil@client:/users/emil/sledge-client/tests/
+
+
+# lab-dell (don't forget to provide the private key in the config file inside .ssh folder)
+# rsync -ru --progress --exclude={'thirdparty','res-*','err.dat','out*','*.log'} ./tests ./runtime lab@161.253.75.227:/home/lab/sledge-emil/
+
+# CMU (don't forget to provide the private key in the config file inside .ssh folder)
+# rsync -ru --progress --exclude={'thirdparty','res','err.dat','out*','*.log'} ./tests ./runtime gwu@arena0.andrew.cmu.edu:/home/gwu/sledge/
+
+# esma
+# rsync -ru --progress --exclude={'thirdparty','res-*','err.dat','out*','*.log'} ./tests emil@161.253.75.224:/home/emil/sledge-client/

runtime/Makefile

@@ -14,10 +14,10 @@ CFLAGS=-std=c18 -pthread
 CFLAGS+=-D_GNU_SOURCE
 
 # Release Flags
-CFLAGS+=-O3 -flto
+# CFLAGS+=-O3 -flto
 
 # Debugging Flags
-# CFLAGS+=-O0 -g3
+CFLAGS+=-O0 -g3
 
 # CFI Sanitizer
 # CFLAGS+=-fvisibility=default -fsanitize=cfi
@@ -43,6 +43,7 @@ CFLAGS += -DEXECUTION_HISTOGRAM
 
 # It is recommended (not mandatory) to enable this flag along with the EXECUTION_HISTOGRAM flag:
 # CFLAGS += -DADMISSIONS_CONTROL
+# CFLAGS += -DTRAFFIC_CONTROL
 
 # Debugging Flags
 
@@ -62,6 +63,7 @@ CFLAGS += -DEXECUTION_HISTOGRAM
 # Various Informational Logs for Debugging
 # CFLAGS += -DLOG_EXECUTION_HISTOGRAM
 # CFLAGS += -DLOG_ADMISSIONS_CONTROL
+# CFLAGS += -DLOG_TRAFFIC_CONTROL
 # CFLAGS += -DLOG_CONTEXT_SWITCHES
 # CFLAGS += -DLOG_HTTP_PARSER
 # CFLAGS += -DLOG_TENANT_LOADING
@@ -74,7 +76,7 @@ CFLAGS += -DEXECUTION_HISTOGRAM
 
 # This adds an array of sandbox states to all sandbox structs and appends states at each transition
 # The history trucates when the number of elements equal SANDBOX_STATE_HISTORY_CAPACITY 
-# CFLAGS += -DLOG_STATE_CHANGES
+CFLAGS += -DLOG_STATE_CHANGES
 
 # This dumps per module *.csv files containing the cycle a sandbox has been in RUNNING when each
 # page is allocated. This helps understand the relationship to memory allocation and execution time.

runtime/include/arch/x86_64/context.h

@@ -56,7 +56,9 @@ arch_context_restore_fast(mcontext_t *active_context, struct arch_context *sandb
 	assert(sandbox_context != NULL);
 
 	/* Assumption: Base Context is only ever used by arch_context_switch */
-	assert(sandbox_context != &worker_thread_base_context);
+	// assert(sandbox_context != &worker_thread_base_context);
+	/* Assumption: Not switching to the same context */
+	assert(active_context != &sandbox_context->mctx);
 
 	assert(sandbox_context->regs[UREG_SP]);
 	assert(sandbox_context->regs[UREG_IP]);

runtime/include/current_sandbox.h

@@ -10,6 +10,10 @@
 extern thread_local struct sandbox *worker_thread_current_sandbox;
 
 void current_sandbox_start(void);
+void current_sandbox_exit(void);
+void interrupted_sandbox_exit(void);
+int  sandbox_validate_self_lifetime(struct sandbox *);
+void sandbox_kill_self(struct sandbox *);
 
 /**
  * Getter for the current sandbox executing on this thread
@@ -51,16 +55,27 @@ current_sandbox_set(struct sandbox *sandbox)
 		/* This is because the event core does not maintain core-assigned deadline */
 		if (!listener_thread_is_running()) runtime_worker_threads_deadline[worker_thread_idx] = UINT64_MAX;
 	} else {
+		assert(sandbox->state == SANDBOX_RUNNABLE || sandbox->state == SANDBOX_PREEMPTED);
+		// if(sandbox->state == SANDBOX_PREEMPTED && sandbox->original_owner_worker_idx != worker_thread_idx) printf("SAND_id: %lu, WASM_id: %lu, wrk: %d\n", sandbox->id, sledge_abi__current_wasm_module_instance.abi.memory.id, worker_thread_idx);
+		// else 
+		// printf("Else SAND_id: %lu, WASM_id: %lu, wrk: %d, orig_wrk %d\n", sandbox->id, sledge_abi__current_wasm_module_instance.abi.memory.id, worker_thread_idx, sandbox->original_owner_worker_idx);
 		sledge_abi__current_wasm_module_instance.wasi_context = sandbox->wasi_context;
+		assert(sandbox->memory->abi.capacity > 0);
+		assert(sandbox->memory->abi.size > 0);
+		assert(sandbox->memory->abi.max > 0);
 		memcpy(&sledge_abi__current_wasm_module_instance.abi.memory, &sandbox->memory->abi,
 		       sizeof(struct sledge_abi__wasm_memory));
+assert(sledge_abi__current_wasm_module_instance.abi.memory.size == sandbox->memory->abi.size);
+assert(sledge_abi__current_wasm_module_instance.abi.memory.id == sandbox->id);
 		sledge_abi__current_wasm_module_instance.abi.table = sandbox->module->indirect_table;
 		wasm_globals_update_if_used(&sandbox->globals, 0,
 		                            &sledge_abi__current_wasm_module_instance.abi.wasmg_0);
 		worker_thread_current_sandbox = sandbox;
+		
 		if (!listener_thread_is_running())
 			runtime_worker_threads_deadline[worker_thread_idx] = sandbox->absolute_deadline;
 	}
+	barrier();
 }
 
 extern void current_sandbox_sleep();

runtime/include/dbf.h

@@ -0,0 +1,75 @@
+#pragma once
+
+#include <stdlib.h>
+#include "tenant.h"
+#include "message.h"
+
+#define DBF_USE_LINKEDLIST
+// static const bool USING_AGGREGATED_GLOBAL_DBF = true;
+
+/* Returns pointer back if successful, null otherwise */
+// extern void *global_dbf;
+extern void **global_virt_worker_dbfs;
+extern void *global_worker_dbf;
+
+struct demand_node {
+	struct ps_list list;
+	uint64_t abs_deadline;
+	uint64_t demand;
+	// uint64_t demand_sum;
+	// struct sandbox_metadata *sandbox_meta;
+	struct tenant *tenant;
+};
+
+typedef enum dbf_update_mode
+{
+	DBF_CHECK_AND_ADD_DEMAND,               /* normal mode for adding new sandbox demands */
+	DBF_FORCE_ADD_NEW_SANDBOX_DEMAND,       /* work-conservation mode*/
+	DBF_FORCE_ADD_MANUAL_DEMAND,            /* work-conservation mode*/
+	DBF_REDUCE_EXISTING_DEMAND,             /* normal mode for reducing existing sandbox demands */
+	// DBF_CHECK_EXISTING_SANDBOX_EXTRA_DEMAND, /* special case when a sandbox goes over its expected exec */
+	DBF_DELETE_EXISTING_DEMAND             /* normal mode for removing existing sandbox demand */
+} dbf_update_mode_t;
+
+typedef int (*dbf_get_worker_idx_fn_t)(void *);
+typedef uint64_t (*dbf_get_time_of_oversupply_fn_t)(void *);
+typedef void (*dbf_print_fn_t)(void *, uint64_t);
+typedef bool (*dbf_try_update_demand_fn_t)(void *, uint64_t, uint64_t, uint64_t, uint64_t, dbf_update_mode_t, void *, struct sandbox_metadata *sandbox_meta);
+typedef uint64_t (*dbf_get_demand_overgone_its_supply_at_fn_t)(void *, uint64_t, uint64_t, uint64_t);
+typedef void (*dbf_free_fn_t)(void *);
+
+struct dbf_config {
+	dbf_get_worker_idx_fn_t                    get_worker_idx_fn;
+	// dbf_get_max_relative_dl_fn_t               get_max_relative_dl_fn;
+	dbf_get_time_of_oversupply_fn_t            get_time_of_oversupply_fn;
+	dbf_print_fn_t                             print_fn;
+	// dbf_grow_fn_t                              grow_fn;
+	dbf_try_update_demand_fn_t                 try_update_demand_fn;
+	dbf_get_demand_overgone_its_supply_at_fn_t get_demand_overgone_its_supply_at_fn;
+	dbf_free_fn_t                              free_fn;
+};
+
+int      dbf_get_worker_idx(void *);
+// uint64_t dbf_get_max_relative_dl(void *);
+uint64_t dbf_get_time_of_oversupply(void *);
+void     dbf_print(void *, uint64_t);
+// void    *dbf_grow(void *, uint64_t);
+bool     dbf_try_update_demand(void *, uint64_t, uint64_t, uint64_t, uint64_t, dbf_update_mode_t, void *, struct sandbox_metadata *sandbox_meta);
+uint64_t dbf_get_demand_overgone_its_supply_at(void *, uint64_t, uint64_t, uint64_t);
+void     dbf_free(void *);
+
+void dbf_plug_functions(struct dbf_config *config);
+
+void *dbf_list_initialize(uint32_t, uint8_t, int, struct tenant *);
+void *dbf_array_initialize(uint32_t, uint8_t, int, struct tenant *);
+void *dbf_initialize(uint32_t num_of_workers, uint8_t reservation_percentile, int worker_idx, struct tenant *tenant);
+
+
+bool
+dbf_list_try_add_new_demand(void *dbf_raw, uint64_t start_time, uint64_t abs_deadline, uint64_t adjustment, struct sandbox_metadata *sm);
+
+void
+dbf_list_force_add_extra_slack(void *dbf_raw, struct sandbox_metadata *sm, uint64_t adjustment);
+
+void
+dbf_list_reduce_demand(struct sandbox_metadata *sm, uint64_t adjustment, bool delete_node);

runtime/include/global_request_scheduler.h

@@ -3,6 +3,7 @@
 #include <stdint.h>
 
 #include "sandbox_types.h"
+#include "sandbox_functions.h"
 
 /* Returns pointer back if successful, null otherwise */
 typedef struct sandbox *(*global_request_scheduler_add_fn_t)(struct sandbox *);
@@ -18,8 +19,13 @@ struct global_request_scheduler_config {
 };
 
 
-void            global_request_scheduler_initialize(struct global_request_scheduler_config *config);
-struct sandbox *global_request_scheduler_add(struct sandbox *);
-int             global_request_scheduler_remove(struct sandbox **);
-int             global_request_scheduler_remove_if_earlier(struct sandbox **, uint64_t targed_deadline);
-uint64_t        global_request_scheduler_peek(void);
+void                    global_request_scheduler_initialize(struct global_request_scheduler_config *config);
+struct sandbox         *global_request_scheduler_add(struct sandbox *);
+int                     global_request_scheduler_remove(struct sandbox **);
+int                     global_request_scheduler_remove_if_earlier(struct sandbox **, uint64_t targed_deadline);
+uint64_t                global_request_scheduler_peek(void);
+void                    global_request_scheduler_update_highest_priority(const void *element);
+struct sandbox_metadata global_request_scheduler_peek_metadata();
+// struct sandbox_metadata global_request_scheduler_peek_metadata_must_lock(const uint64_t now, struct sandbox *);
+void                    global_default_update_highest_priority(const void *element);
+struct sandbox_metadata global_default_peek_metadata();

runtime/include/global_request_scheduler_mtdbf.h

@@ -0,0 +1,5 @@
+#pragma once
+
+#include "global_request_scheduler.h"
+
+void global_request_scheduler_mtdbf_initialize();

runtime/include/http.h

@@ -36,6 +36,20 @@
 	"Connection: close\r\n"
 #define HTTP_RESPONSE_404_NOT_FOUND_LENGTH 59
 
+#define HTTP_RESPONSE_408_REQUEST_TIMEOUT  \
+	"HTTP/1.1 408 Request Timeout\r\n" \
+	"Server: SLEdge\r\n"               \
+	"Connection: close\r\n"            \
+	"\r\n"
+#define HTTP_RESPONSE_408_REQUEST_TIMEOUT_LENGTH 67
+
+#define HTTP_RESPONSE_409_CONFLICT  \
+	"HTTP/1.1 409 Conflict\r\n" \
+	"Server: SLEdge\r\n"        \
+	"Connection: close\r\n"     \
+	"\r\n"
+#define HTTP_RESPONSE_409_CONFLICT_LENGTH 60
+
 #define HTTP_RESPONSE_413_PAYLOAD_TOO_LARGE  \
 	"HTTP/1.1 413 Payload Too Large\r\n" \
 	"Server: SLEdge\r\n"                 \
@@ -70,6 +84,10 @@ http_header_build(int status_code)
 		return HTTP_RESPONSE_400_BAD_REQUEST;
 	case 404:
 		return HTTP_RESPONSE_404_NOT_FOUND;
+	case 408:
+		return HTTP_RESPONSE_408_REQUEST_TIMEOUT;
+	case 409:
+		return HTTP_RESPONSE_409_CONFLICT;
 	case 413:
 		return HTTP_RESPONSE_413_PAYLOAD_TOO_LARGE;
 	case 429:
@@ -91,6 +109,10 @@ http_header_len(int status_code)
 		return HTTP_RESPONSE_400_BAD_REQUEST_LENGTH;
 	case 404:
 		return HTTP_RESPONSE_404_NOT_FOUND_LENGTH;
+	case 408:
+		return HTTP_RESPONSE_408_REQUEST_TIMEOUT_LENGTH;
+	case 409:
+		return HTTP_RESPONSE_409_CONFLICT_LENGTH;
 	case 413:
 		return HTTP_RESPONSE_413_PAYLOAD_TOO_LARGE_LENGTH;
 	case 429:

runtime/include/http_router.h

@@ -49,10 +49,7 @@ http_router_add_route(http_router_t *router, struct route_config *config, struct
 #endif
 
 	const uint64_t expected_execution = route.relative_deadline / 2;
-#ifdef ADMISSIONS_CONTROL
-	/* Addmissions Control setup */
 	route.execution_histogram.estimated_execution = expected_execution;
-#endif
 
 #ifdef EXECUTION_HISTOGRAM
 	/* Execution Histogram setup */

runtime/include/listener_thread.h

@@ -5,10 +5,26 @@
 
 #include "http_session.h"
 #include "module.h"
+#include "ck_ring.h"
+#include "sandbox_state.h"
+#include "dbf.h"
 
-#define LISTENER_THREAD_CORE_ID 1
+#define LISTENER_THREAD_CORE_ID   1
+#define LISTENER_THREAD_RING_SIZE 10240 /* the acutal size becomes 255 */
 
-extern pthread_t listener_thread_id;
+struct comm_with_worker {
+	ck_ring_t      worker_ring;
+	struct message worker_ring_buffer[LISTENER_THREAD_RING_SIZE];
+	int            worker_idx;
+}; // __attribute__((aligned(CACHE_PAD))); ///// TODO: this necessary?
+
+CK_RING_PROTOTYPE(message, message)
+
+extern pthread_t                listener_thread_id;
+extern int                      listener_thread_epoll_file_descriptor;
+extern struct comm_with_worker *comm_from_workers;
+extern struct comm_with_worker *comm_from_workers_extra;
+extern struct comm_with_worker *comm_to_workers;
 
 void           listener_thread_initialize(void);
 noreturn void *listener_thread_main(void *dummy);
@@ -23,3 +39,14 @@ listener_thread_is_running()
 {
 	return pthread_self() == listener_thread_id;
 }
+
+static inline void
+comm_with_workers_init(struct comm_with_worker *comm_with_workers)
+{
+	assert(comm_with_workers);
+
+	for (int worker_idx = 0; worker_idx < runtime_worker_threads_count; worker_idx++) {
+		ck_ring_init(&comm_with_workers[worker_idx].worker_ring, LISTENER_THREAD_RING_SIZE);
+		comm_with_workers[worker_idx].worker_idx = worker_idx;
+	}
+}

runtime/include/local_runqueue.h

@@ -17,8 +17,10 @@ struct local_runqueue_config {
 	local_runqueue_get_next_fn_t get_next_fn;
 };
 
-void            local_runqueue_add(struct sandbox *);
-void            local_runqueue_delete(struct sandbox *);
-bool            local_runqueue_is_empty();
-struct sandbox *local_runqueue_get_next();
-void            local_runqueue_initialize(struct local_runqueue_config *config);
+void                    local_runqueue_add(struct sandbox *);
+void                    local_runqueue_delete(struct sandbox *);
+bool                    local_runqueue_is_empty();
+struct sandbox         *local_runqueue_get_next();
+void                    local_runqueue_initialize(struct local_runqueue_config *config);
+// void                    local_runqueue_update_highest_priority(const void *element);
+// struct sandbox_metadata local_runqueue_peek_metadata();

runtime/include/local_runqueue_mtdbf.h

@@ -0,0 +1,7 @@
+#pragma once
+
+#include "module.h"
+
+void local_runqueue_mtdbf_initialize();
+
+size_t queue_length();

runtime/include/message.h

@@ -0,0 +1,33 @@
+#ifndef MESSAGE_H
+#define MESSAGE_H
+
+#include "runtime.h"
+
+typedef enum
+{
+	MESSAGE_CFW_PULLED_NEW_SANDBOX,
+	MESSAGE_CFW_REDUCE_DEMAND, 
+	MESSAGE_CFW_DELETE_SANDBOX, /* normal mode for deleting new sandbox demands */
+	MESSAGE_CFW_EXTRA_DEMAND_REQUEST,
+	MESSAGE_CFW_WRITEBACK_PREEMPTION,
+	MESSAGE_CFW_WRITEBACK_OVERSHOOT,
+
+	MESSAGE_CTW_SHED_CURRENT_JOB
+} message_type_t;
+
+struct message {
+	uint64_t                 sandbox_id;
+	uint64_t                 adjustment;
+	uint64_t                 total_running_duration;
+	uint64_t                 remaining_exec;
+	uint64_t                 timestamp;
+	struct sandbox          *sandbox;
+	struct sandbox_metadata *sandbox_meta;
+	message_type_t           message_type;
+	int                      sender_worker_idx;
+	uint8_t                  state;
+	bool                     exceeded_estimation;
+}; // PAGE_ALIGNED;
+
+
+#endif /* MESSAGE_H */

runtime/include/module.h

@@ -1,5 +1,10 @@
 #pragma once
 
+// #include <string.h>
+// #include <sys/socket.h>
+// #include <sys/types.h>
+// #include <netdb.h>
+
 #include "current_wasm_module_instance.h"
 #include "pool.h"
 #include "sledge_abi_symbols.h"
@@ -113,9 +118,11 @@ module_initialize_pools(struct module *module)
 {
 	/* Create only a single pool for the preprocessing module, since it is executed only by the event core. */
 	const int n = module->type == APP_MODULE ? runtime_worker_threads_count : 1;
-	for (int i = 0; i < n; i++) {
-		wasm_memory_pool_init(&module->pools[i].memory, false);
-		wasm_stack_pool_init(&module->pools[i].stack, false);
+	for (int i = 0; i < 1; i++) {
+		// wasm_memory_pool_init(&module->pools[i].memory, false);
+		// wasm_stack_pool_init(&module->pools[i].stack, false);
+		wasm_memory_pool_init(&module->pools[i].memory, true);
+		wasm_stack_pool_init(&module->pools[i].stack, true);
 	}
 }
 
@@ -123,7 +130,7 @@ static inline void
 module_deinitialize_pools(struct module *module)
 {
 	const int n = module->type == APP_MODULE ? runtime_worker_threads_count : 1;
-	for (int i = 0; i < n; i++) {
+	for (int i = 0; i < 1; i++) {
 		wasm_memory_pool_deinit(&module->pools[i].memory);
 		wasm_stack_pool_deinit(&module->pools[i].stack);
 	}
@@ -167,7 +174,8 @@ module_allocate_stack(struct module *module)
 {
 	assert(module != NULL);
 
-	struct wasm_stack *stack = wasm_stack_pool_remove_nolock(&module->pools[worker_thread_idx].stack);
+	// struct wasm_stack *stack = wasm_stack_pool_remove_nolock(&module->pools[worker_thread_idx].stack);
+	struct wasm_stack *stack = wasm_stack_pool_remove(&module->pools[0].stack);
 
 	if (stack == NULL) {
 		stack = wasm_stack_alloc(module->stack_size);
@@ -178,10 +186,17 @@ module_allocate_stack(struct module *module)
 }
 
 static inline void
-module_free_stack(struct module *module, struct wasm_stack *stack)
+module_free_stack(struct module *module, struct wasm_stack *stack, int orig_wrk_idx)
 {
+	assert(orig_wrk_idx >= 0);
+	// if (module->pools[worker_thread_idx].stack.size == RUNTIME_WORKER_POOL_SIZE) {
+	// 	assert(0);
+	// 	wasm_stack_free(stack);
+	// 	return;
+	// }
 	wasm_stack_reinit(stack);
-	wasm_stack_pool_add_nolock(&module->pools[worker_thread_idx].stack, stack);
+	// wasm_stack_pool_add_nolock(&module->pools[worker_thread_idx].stack, stack);
+	wasm_stack_pool_add(&module->pools[0].stack, stack);
 }
 
 static inline struct wasm_memory *
@@ -198,7 +213,8 @@ module_allocate_linear_memory(struct module *module)
 	assert(starting_bytes <= (uint64_t)UINT32_MAX + 1);
 	assert(max_bytes <= (uint64_t)UINT32_MAX + 1);
 
-	struct wasm_memory *linear_memory = wasm_memory_pool_remove_nolock(&module->pools[worker_thread_idx].memory);
+	// struct wasm_memory *linear_memory = wasm_memory_pool_remove_nolock(&module->pools[worker_thread_idx].memory);
+	struct wasm_memory *linear_memory = wasm_memory_pool_remove(&module->pools[0].memory);
 	if (linear_memory == NULL) {
 		linear_memory = wasm_memory_alloc(starting_bytes, max_bytes);
 		if (unlikely(linear_memory == NULL)) return NULL;
@@ -207,9 +223,26 @@ module_allocate_linear_memory(struct module *module)
 	return linear_memory;
 }
 
+thread_local static uint16_t max_size = 0;
+
 static inline void
-module_free_linear_memory(struct module *module, struct wasm_memory *memory)
+module_free_linear_memory(struct module *module, struct wasm_memory *memory, int orig_wrk_idx)
 {
+	assert(orig_wrk_idx >= 0);
+	// if (module->pools[worker_thread_idx].memory.size == RUNTIME_WORKER_POOL_SIZE) {
+	// 	assert(0);
+	// 	wasm_memory_free(memory);
+	// 	return;
+	// }
 	wasm_memory_reinit(memory, module->abi.starting_pages * WASM_PAGE_SIZE);
-	wasm_memory_pool_add_nolock(&module->pools[worker_thread_idx].memory, memory);
+	// wasm_memory_pool_add_nolock(&module->pools[worker_thread_idx].memory, memory);
+	wasm_memory_pool_add(&module->pools[0].memory, memory);
+	// if (17 < module->pools[0].memory.size) {
+	// 	// max_size = module->pools[0].memory.size;
+	// 	printf("Module %s - [Worker]=Sizes\n",  module->path);
+	// 	for (int i=0; i< 1; i++) {
+	// 		printf("%u ", module->pools[i].memory.size);
+	// 	}
+	// 	printf("\n\n");
+	// }
 }

runtime/include/pool.h

@@ -12,6 +12,7 @@
 		bool                use_lock;                                                                      \
 		lock_t              lock;                                                                          \
 		struct ps_list_head list;                                                                          \
+		uint16_t            size;                                                                          \
 	};                                                                                                         \
                                                                                                                    \
 	static inline bool STRUCT_NAME##_pool_is_empty(struct STRUCT_NAME##_pool *self)                            \
@@ -26,6 +27,7 @@
 		ps_list_head_init(&self->list);                                                                    \
 		self->use_lock = use_lock;                                                                         \
 		if (use_lock) lock_init(&self->lock);                                                              \
+		self->size = 0;                                                                                    \
 	}                                                                                                          \
                                                                                                                    \
 	static inline void STRUCT_NAME##_pool_deinit(struct STRUCT_NAME##_pool *self)                              \
@@ -52,6 +54,8 @@
 		obj = ps_list_head_first_d(&self->list, struct STRUCT_NAME);                                       \
 		assert(obj);                                                                                       \
 		ps_list_rem_d(obj);                                                                                \
+		assert(self->size > 0);                                                                            \
+		self->size--;                                                                                      \
                                                                                                                    \
 		return obj;                                                                                        \
 	}                                                                                                          \
@@ -79,6 +83,8 @@
 		assert(!self->use_lock || lock_is_locked(&self->lock));                                            \
                                                                                                                    \
 		ps_list_head_add_d(&self->list, obj);                                                              \
+		self->size++;                                                                                      \
+		/*assert(self->size <= RUNTIME_WORKER_POOL_SIZE); */                                                       \
 	}                                                                                                          \
                                                                                                                    \
 	static inline void STRUCT_NAME##_pool_add(struct STRUCT_NAME##_pool *self, struct STRUCT_NAME *obj)        \

runtime/include/priority_queue.h

@@ -18,16 +18,20 @@
  * @returns priority (a uint64_t)
  */
 typedef uint64_t (*priority_queue_get_priority_fn_t)(void *element);
+typedef void (*priority_queue_update_priority_fn_t)(const void *element);
+typedef void (*priority_queue_update_idx_fn_t)(void *element, size_t idx);
 
 /* We assume that priority is expressed in terms of a 64 bit unsigned integral */
 struct priority_queue {
-	priority_queue_get_priority_fn_t get_priority_fn;
-	bool                             use_lock;
-	lock_t                           lock;
-	uint64_t                         highest_priority;
-	size_t                           size;
-	size_t                           capacity;
-	void                            *items[];
+	priority_queue_get_priority_fn_t    get_priority_fn;
+	priority_queue_update_priority_fn_t update_priority_fn;
+	priority_queue_update_idx_fn_t      update_idx_fn;
+	bool                                use_lock;
+	lock_t                              lock;
+	uint64_t                            highest_priority;
+	size_t                              size;
+	size_t                              capacity;
+	void	                       *items[];
 };
 
 /**
@@ -47,6 +51,11 @@ static inline void
 priority_queue_update_highest_priority(struct priority_queue *priority_queue, const uint64_t priority)
 {
 	priority_queue->highest_priority = priority;
+
+	if (priority_queue->update_priority_fn) {
+		// priority_queue->update_priority_fn((priority_queue->size > 0) ? priority_queue->items[1] : NULL);
+		priority_queue->update_priority_fn(priority_queue->items[1]);
+	}
 }
 
 /**
@@ -60,6 +69,7 @@ priority_queue_append(struct priority_queue *priority_queue, void *new_item)
 {
 	assert(priority_queue != NULL);
 	assert(new_item != NULL);
+	// assert(priority_queue->update_idx_fn != NULL);
 	assert(!priority_queue->use_lock || lock_is_locked(&priority_queue->lock));
 
 	int rc;
@@ -67,6 +77,9 @@ priority_queue_append(struct priority_queue *priority_queue, void *new_item)
 	if (unlikely(priority_queue->size > priority_queue->capacity)) panic("PQ overflow");
 	if (unlikely(priority_queue->size == priority_queue->capacity)) goto err_enospc;
 	priority_queue->items[++priority_queue->size] = new_item;
+	if (priority_queue->update_idx_fn) {
+		priority_queue->update_idx_fn(priority_queue->items[priority_queue->size], priority_queue->size);
+	}
 
 	rc = 0;
 done:
@@ -116,6 +129,10 @@ priority_queue_percolate_up(struct priority_queue *priority_queue)
 		void *temp                   = priority_queue->items[i / 2];
 		priority_queue->items[i / 2] = priority_queue->items[i];
 		priority_queue->items[i]     = temp;
+		if (priority_queue->update_idx_fn) {
+			priority_queue->update_idx_fn(priority_queue->items[i], i);
+			priority_queue->update_idx_fn(priority_queue->items[i / 2], i / 2);
+		}
 		/* If percolated to highest priority, update highest priority */
 		if (i / 2 == 1)
 			priority_queue_update_highest_priority(priority_queue, priority_queue->get_priority_fn(
@@ -168,7 +185,7 @@ priority_queue_percolate_down(struct priority_queue *priority_queue, int parent_
 	assert(priority_queue != NULL);
 	assert(priority_queue->get_priority_fn != NULL);
 	assert(!priority_queue->use_lock || lock_is_locked(&priority_queue->lock));
-	assert(!listener_thread_is_running());
+	// assert(!listener_thread_is_running());
 
 	bool update_highest_value = parent_index == 1;
 
@@ -183,6 +200,11 @@ priority_queue_percolate_down(struct priority_queue *priority_queue, int parent_
 		void *temp                                  = priority_queue->items[smallest_child_index];
 		priority_queue->items[smallest_child_index] = priority_queue->items[parent_index];
 		priority_queue->items[parent_index]         = temp;
+		if (priority_queue->update_idx_fn) {
+			priority_queue->update_idx_fn(priority_queue->items[smallest_child_index],
+			                              smallest_child_index);
+			priority_queue->update_idx_fn(priority_queue->items[parent_index], parent_index);
+		}
 
 		parent_index     = smallest_child_index;
 		left_child_index = 2 * parent_index;
@@ -216,8 +238,8 @@ priority_queue_dequeue_if_earlier_nolock(struct priority_queue *priority_queue,
 	assert(priority_queue != NULL);
 	assert(dequeued_element != NULL);
 	assert(priority_queue->get_priority_fn != NULL);
-	assert(!listener_thread_is_running());
 	assert(!priority_queue->use_lock || lock_is_locked(&priority_queue->lock));
+	// assert(!listener_thread_is_running());
 
 	int return_code;
 
@@ -225,9 +247,13 @@ priority_queue_dequeue_if_earlier_nolock(struct priority_queue *priority_queue,
 	if (priority_queue_is_empty(priority_queue) || priority_queue->highest_priority >= target_deadline)
 		goto err_enoent;
 
-	*dequeued_element                             = priority_queue->items[1];
+	*dequeued_element = priority_queue->items[1];
+	if (priority_queue->update_idx_fn) { priority_queue->update_idx_fn(*dequeued_element, 0); }
 	priority_queue->items[1]                      = priority_queue->items[priority_queue->size];
 	priority_queue->items[priority_queue->size--] = NULL;
+	if (priority_queue->update_idx_fn && priority_queue->items[1]) {
+		priority_queue->update_idx_fn(priority_queue->items[1], 1);
+	}
 
 	priority_queue_percolate_down(priority_queue, 1);
 	return_code = 0;
@@ -287,6 +313,29 @@ priority_queue_initialize(size_t capacity, bool use_lock, priority_queue_get_pri
 	return priority_queue;
 }
 
+/**
+ * Initializes some additional properties of Priority Queue Data structure. TODO: Unite with above
+ * @param capacity the number of elements to store in the data structure
+ * @param use_lock indicates that we want a concurrent data structure
+ * @param get_priority_fn pointer to a function that returns the priority of an element
+ * @return priority queue
+ */
+static inline struct priority_queue *
+priority_queue_initialize_new(size_t capacity, bool use_lock, priority_queue_get_priority_fn_t get_priority_fn,
+                              priority_queue_update_priority_fn_t update_priority_fn,
+                              priority_queue_update_idx_fn_t      update_idx_fn)
+{
+	assert(update_idx_fn != NULL);
+
+	struct priority_queue *priority_queue = priority_queue_initialize(capacity, use_lock, get_priority_fn);
+
+	priority_queue->update_priority_fn = update_priority_fn;
+	priority_queue->update_idx_fn      = update_idx_fn;
+	if (update_priority_fn) priority_queue->update_priority_fn(NULL);
+
+	return priority_queue;
+}
+
 /**
  * Double capacity of priority queue
  * Note: currently there is no equivalent call for PQs that are not thread-local and need to be locked because it is
@@ -412,6 +461,10 @@ priority_queue_delete_nolock(struct priority_queue *priority_queue, void *value)
 		if (priority_queue->items[i] == value) {
 			priority_queue->items[i]                      = priority_queue->items[priority_queue->size];
 			priority_queue->items[priority_queue->size--] = NULL;
+			if (priority_queue->update_idx_fn) {
+				priority_queue->update_idx_fn(value, 0);
+				priority_queue->update_idx_fn(priority_queue->items[i], i);
+			}
 			priority_queue_percolate_down(priority_queue, i);
 			return 0;
 		}
@@ -438,6 +491,52 @@ priority_queue_delete(struct priority_queue *priority_queue, void *value)
 	return rc;
 }
 
+/**
+ * @param priority_queue - the priority queue we want to delete from
+ * @param idx - the index of the value we want to delete
+ */
+static inline void
+priority_queue_delete_by_idx_nolock(struct priority_queue *priority_queue, const void *value_to_remove,
+                                    const size_t idx)
+{
+	assert(priority_queue != NULL);
+	assert(idx <= priority_queue->size);
+	assert(idx >= 1);
+	assert(priority_queue->update_idx_fn);
+	if (scheduler != SCHEDULER_MTDS && scheduler != SCHEDULER_MTDBF) assert(!listener_thread_is_running());
+	assert(!priority_queue->use_lock || lock_is_locked(&priority_queue->lock));
+
+	void *r = priority_queue->items[idx];
+	assert(r == value_to_remove);
+	priority_queue->update_idx_fn(r, 0);
+
+	if (idx == priority_queue->size) {
+		// priority_queue->size--;
+		priority_queue->items[priority_queue->size--] = NULL;
+		priority_queue_percolate_down(priority_queue, idx);
+		return;
+	}
+
+	priority_queue->items[idx]                    = priority_queue->items[priority_queue->size];
+	priority_queue->items[priority_queue->size--] = NULL;
+	priority_queue->update_idx_fn(priority_queue->items[idx], idx);
+	priority_queue_percolate_down(priority_queue, idx);
+}
+
+/**
+ * @param priority_queue - the priority queue we want to delete from
+ * @param idx - the index of the value we want to delete
+ * @returns 0 on success. -1 on not found
+ */
+static inline void
+priority_queue_delete_by_idx(struct priority_queue *priority_queue, const void *value_to_remove, const size_t idx)
+{
+	lock_node_t node = {};
+	lock_lock(&priority_queue->lock, &node);
+	priority_queue_delete_by_idx_nolock(priority_queue, value_to_remove, idx);
+	lock_unlock(&priority_queue->lock, &node);
+}
+
 /**
  * @param priority_queue - the priority queue we want to add to
  * @param dequeued_element a pointer to set to the dequeued element

runtime/include/ps_list.h

@@ -183,7 +183,13 @@ ps_list_ll_rem(struct ps_list *l)
 	for (iter   = ps_list_head_first((head), __typeof__(*iter), lname); !ps_list_is_head((head), iter, lname); \
 	     (iter) = ps_list_next(iter, lname))
 
+/* Reverse iteration without mutating the list */
+#define ps_list_foreach_rev(head, iter, lname)                                                                         \
+	for (iter   = ps_list_head_last((head), __typeof__(*iter), lname); !ps_list_is_head((head), iter, lname); \
+	     (iter) = ps_list_prev(iter, lname))
+
 #define ps_list_foreach_d(head, iter) ps_list_foreach(head, iter, PS_LIST_DEF_NAME)
+#define ps_list_foreach_rev_d(head, iter) ps_list_foreach_rev(head, iter, PS_LIST_DEF_NAME)
 
 /*
  * Iteration where the current node can be ps_list_rem'ed.

runtime/include/route.h

@@ -4,6 +4,9 @@
 #include <stdint.h>
 
 #include "execution_histogram.h"
+// #include "admissions_info.h"
+// #include "estimated_exec_info.h"
+// #include "module.h"
 #include "http_route_total.h"
 #include "module.h"
 #include "perf_window.h"

runtime/include/route_config.h

@@ -5,6 +5,7 @@
 #include <stdlib.h>
 
 #include "admissions_control.h"
+#include "traffic_control.h"
 #include "runtime.h"
 #include "scheduler_options.h"
 
@@ -95,7 +96,7 @@ route_config_validate(struct route_config *config, bool *did_set)
 			return -1;
 		}
 
-		if (config->relative_deadline_us > (uint32_t)RUNTIME_RELATIVE_DEADLINE_US_MAX) {
+		if (config->relative_deadline_us == 0 || config->relative_deadline_us > (uint32_t)RUNTIME_RELATIVE_DEADLINE_US_MAX) {
 			fprintf(stderr, "Relative-deadline-us must be between 0 and %u, was %u\n",
 			        (uint32_t)RUNTIME_RELATIVE_DEADLINE_US_MAX, config->relative_deadline_us);
 			return -1;

runtime/include/runtime.h

@@ -24,10 +24,14 @@
 
 #define RUNTIME_LOG_FILE                 "sledge.log"
 #define RUNTIME_MAX_EPOLL_EVENTS         128
-#define RUNTIME_MAX_TENANT_COUNT         32
+#define RUNTIME_MAX_TENANT_COUNT         320
 #define RUNTIME_RELATIVE_DEADLINE_US_MAX 3600000000 /* One Hour. Fits in uint32_t */
-#define RUNTIME_RUNQUEUE_SIZE            256        /* Minimum guaranteed size. Might grow! */
+#define RUNTIME_RUNQUEUE_SIZE            2560        /* Minimum guaranteed size. Might grow! */
 #define RUNTIME_TENANT_QUEUE_SIZE        4096
+#define RUNTIME_WORKER_POOL_SIZE         8 /* Set to zero if no pooling desired */
+
+#define RUNTIME_MAX_CPU_UTIL_PERCENTILE 100
+#define RUNTIME_MAX_ALIVE_SANDBOXES     1048576 // 1 << 20
 
 enum RUNTIME_SIGALRM_HANDLER
 {
@@ -40,16 +44,22 @@ extern bool                         runtime_preemption_enabled;
 extern bool                         runtime_worker_spinloop_pause_enabled;
 extern uint32_t                     runtime_processor_speed_MHz;
 extern uint32_t                     runtime_quantum_us;
+extern uint64_t                     runtime_quantum;
 extern enum RUNTIME_SIGALRM_HANDLER runtime_sigalrm_handler;
 extern pthread_t                   *runtime_worker_threads;
 extern uint32_t                     runtime_worker_threads_count;
 extern int                         *runtime_worker_threads_argument;
 extern uint64_t                    *runtime_worker_threads_deadline;
 extern uint64_t                     runtime_boot_timestamp;
+extern uint64_t                     runtime_max_deadline;
+extern bool                         sandbox_refs[];
+extern uint16_t                     extra_execution_slack_p;
 
 extern void runtime_initialize(void);
 extern void runtime_set_pthread_prio(pthread_t thread, unsigned int nice);
 extern void runtime_set_resource_limits_to_max(void);
+extern void runtime_cleanup();
+void runtime_set_policy_and_prio();
 
 /* External Symbols */
 extern int   expand_memory(void);
@@ -67,3 +77,15 @@ runtime_print_sigalrm_handler(enum RUNTIME_SIGALRM_HANDLER variant)
 		return "TRIAGED";
 	}
 }
+
+static const bool USING_WORK_CONSERVATION = true;
+
+static const bool USING_LOCAL_RUNQUEUE = false;
+static const bool USING_TRY_LOCAL_EXTRA = true;
+
+static const bool USING_WRITEBACK_FOR_PREEMPTION = !USING_LOCAL_RUNQUEUE;
+static const bool USING_WRITEBACK_FOR_OVERSHOOT = false;
+
+static const bool USING_AGGREGATED_GLOBAL_DBF = USING_LOCAL_RUNQUEUE || false;
+
+static const bool USING_EARLIEST_START_FIRST = false;

runtime/include/sandbox_functions.h

@@ -12,12 +12,14 @@
  * Public API              *
  **************************/
 
-struct sandbox *sandbox_alloc(struct module *module, struct http_session *session, struct route *route,
-                              struct tenant *tenant, uint64_t admissions_estimate);
-int             sandbox_prepare_execution_environment(struct sandbox *sandbox);
-void            sandbox_free(struct sandbox *sandbox);
-void            sandbox_main(struct sandbox *sandbox);
-void            sandbox_switch_to(struct sandbox *next_sandbox);
+struct sandbox *sandbox_alloc(struct module *module, struct http_session *session, uint64_t admissions_estimate,
+                              uint64_t sandbox_alloc_timestamp);
+struct sandbox_metadata *sandbox_meta_alloc(struct sandbox *sandbox);
+int                      sandbox_prepare_execution_environment(struct sandbox *sandbox);
+void                     sandbox_free(struct sandbox *sandbox);
+void                     sandbox_main(struct sandbox *sandbox);
+void                     sandbox_switch_to(struct sandbox *next_sandbox);
+void                     sandbox_process_scheduler_updates(struct sandbox *sandbox);
 
 /**
  * Free Linear Memory, leaving stack in place
@@ -28,7 +30,9 @@ sandbox_free_linear_memory(struct sandbox *sandbox)
 {
 	assert(sandbox != NULL);
 	assert(sandbox->memory != NULL);
-	module_free_linear_memory(sandbox->module, (struct wasm_memory *)sandbox->memory);
+	// if (worker_thread_idx != sandbox->original_owner_worker_idx)
+	// 	printf("Me: %d, Orig: %d, Tenant: %s\n", worker_thread_idx, sandbox->original_owner_worker_idx, sandbox->tenant->name);
+	module_free_linear_memory(sandbox->module, (struct wasm_memory *)sandbox->memory, sandbox->original_owner_worker_idx);
 	sandbox->memory = NULL;
 }
 
@@ -48,13 +52,29 @@ static inline uint64_t
 sandbox_get_priority(void *element)
 {
 	struct sandbox *sandbox = (struct sandbox *)element;
+	if (scheduler == SCHEDULER_SJF) return sandbox->remaining_exec;
 	return sandbox->absolute_deadline;
 }
 
+static inline uint64_t
+sandbox_get_priority_global(void *element)
+{
+	struct sandbox *sandbox = (struct sandbox *)element;
+	return sandbox->absolute_deadline - sandbox->remaining_exec;
+}
+
+static inline void
+sandbox_update_pq_idx_in_runqueue(void *element, size_t idx)
+{
+	assert(element);
+	struct sandbox *sandbox     = (struct sandbox *)element;
+	sandbox->pq_idx_in_runqueue = idx;
+}
+
 static inline void
-sandbox_process_scheduler_updates(struct sandbox *sandbox)
+local_sandbox_meta_update_pq_idx_in_tenant_queue(void *element, size_t idx)
 {
-	if (tenant_is_paid(sandbox->tenant)) {
-		atomic_fetch_sub(&sandbox->tenant->remaining_budget, sandbox->last_state_duration);
-	}
+	assert(element);
+	struct sandbox_metadata *sandbox_meta = (struct sandbox_metadata *)element;
+	sandbox_meta->pq_idx_in_tenant_queue  = idx;
 }

runtime/include/sandbox_perf_log.h

@@ -13,9 +13,25 @@ static inline void
 sandbox_perf_log_print_header()
 {
 	if (sandbox_perf_log == NULL) { perror("sandbox perf log"); }
-	fprintf(sandbox_perf_log, "id,tenant,route,state,deadline,actual,queued,uninitialized,allocated,initialized,"
-	                          "runnable,interrupted,preempted,"
-	                          "running_sys,running_user,asleep,returned,complete,error,proc_MHz,payload_size\n");
+	fprintf(sandbox_perf_log,
+	        "id,tenant,route,state,deadline,actual,queued,uninitialized,allocated,initialized,"
+	        "runnable,interrupted,preempted,"
+	        "running_sys,running_user,asleep,returned,complete,error,proc_MHz,response_code,guarantee_type,payload_size\n");
+}
+
+/**
+ * Prints key performance metrics for a denied request (by AC) to perf_log
+ * This is defined by an environment variable
+ * @param module
+ */
+static inline void
+sandbox_perf_log_print_denied_entry(struct tenant *tenant, struct route *route, uint16_t response_code)
+{
+	/* If the log was not defined by an environment variable, early out */
+	if (sandbox_perf_log == NULL) return;
+
+	fprintf(sandbox_perf_log, "-1,%s,%s,Deny,%lu,0,0,0,0,0,0,0,0,0,0,0,0,0,0,%u,%u,0,0\n", tenant->name, route->route,
+	        route->relative_deadline, runtime_processor_speed_MHz, response_code);
 }
 
 /**
@@ -29,16 +45,15 @@ sandbox_perf_log_print_entry(struct sandbox *sandbox)
 	/* If the log was not defined by an environment variable, early out */
 	if (sandbox_perf_log == NULL) return;
 
-	uint64_t queued_duration = sandbox->timestamp_of.dispatched - sandbox->timestamp_of.allocation;
+	uint64_t queued_duration = sandbox->timestamp_of.dispatched - sandbox->timestamp_of.allocation; // TODO: Consider writeback
 
 	/*
 	 * Assumption: A sandbox is never able to free pages. If linear memory management
 	 * becomes more intelligent, then peak linear memory size needs to be tracked
 	 * seperately from current linear memory size.
 	 */
-	fprintf(sandbox_perf_log,
-	        "%lu,%s,%s,%s,%lu,%lu,%lu,%lu,%lu,%lu,%lu,%lu,%lu,%lu,%lu,%lu,%lu,%lu,%lu,%u,%u,%d,%d\n", sandbox->id,
-	        sandbox->tenant->name, sandbox->route->route, sandbox_state_stringify(sandbox->state),
+	fprintf(sandbox_perf_log, "%lu,%s,%s,%s,%lu,%lu,%lu,%lu,%lu,%lu,%lu,%lu,%lu,%lu,%lu,%lu,%lu,%lu,%lu,%u,%u,%d,%d\n",
+	        sandbox->id, sandbox->tenant->name, sandbox->route->route, sandbox_state_stringify(sandbox->state),
 	        sandbox->route->relative_deadline, sandbox->total_time, queued_duration,
 	        sandbox->duration_of_state[SANDBOX_UNINITIALIZED], sandbox->duration_of_state[SANDBOX_ALLOCATED],
 	        sandbox->duration_of_state[SANDBOX_INITIALIZED], sandbox->duration_of_state[SANDBOX_RUNNABLE],
@@ -46,7 +61,7 @@ sandbox_perf_log_print_entry(struct sandbox *sandbox)
 	        sandbox->duration_of_state[SANDBOX_RUNNING_SYS], sandbox->duration_of_state[SANDBOX_RUNNING_USER],
 	        sandbox->duration_of_state[SANDBOX_ASLEEP], sandbox->duration_of_state[SANDBOX_RETURNED],
 	        sandbox->duration_of_state[SANDBOX_COMPLETE], sandbox->duration_of_state[SANDBOX_ERROR],
-	        runtime_processor_speed_MHz, sandbox->response_code, 0, sandbox->payload_size);
+	        runtime_processor_speed_MHz, sandbox->response_code, sandbox->global_queue_type, sandbox->payload_size);
 }
 
 static inline void

runtime/include/sandbox_set_as_allocated.h

@@ -20,11 +20,12 @@ sandbox_set_as_allocated(struct sandbox *sandbox)
 {
 	assert(sandbox);
 	assert(sandbox->state == SANDBOX_UNINITIALIZED);
+	sandbox->state = SANDBOX_ALLOCATED;
 	uint64_t now = __getcycles();
 
 	/* State Change Bookkeeping */
 	assert(now > sandbox->timestamp_of.last_state_change);
-	sandbox->timestamp_of.allocation        = now;
+	// sandbox->timestamp_of.allocation        = now;
 	sandbox->timestamp_of.last_state_change = now;
 	sandbox_state_history_init(&sandbox->state_history);
 	sandbox_state_history_append(&sandbox->state_history, SANDBOX_ALLOCATED);

runtime/include/sandbox_set_as_complete.h

@@ -29,15 +29,12 @@ sandbox_set_as_complete(struct sandbox *sandbox, sandbox_state_t last_state)
 	uint64_t now   = __getcycles();
 
 	switch (last_state) {
-	case SANDBOX_RETURNED: {
-		sandbox->timestamp_of.completion = now;
+	case SANDBOX_RETURNED:
 		break;
-	}
-	default: {
+	default:
 		panic("Sandbox %lu | Illegal transition from %s to Error\n", sandbox->id,
 		      sandbox_state_stringify(last_state));
 	}
-	}
 
 	/* State Change Bookkeeping */
 	assert(now > sandbox->timestamp_of.last_state_change);
@@ -48,13 +45,11 @@ sandbox_set_as_complete(struct sandbox *sandbox, sandbox_state_t last_state)
 	sandbox_state_totals_increment(SANDBOX_COMPLETE);
 	sandbox_state_totals_decrement(last_state);
 
-	struct route *route = sandbox->route;
-
 #ifdef EXECUTION_HISTOGRAM
 	/* Execution Histogram Post Processing */
 	const uint64_t execution_duration = sandbox->duration_of_state[SANDBOX_RUNNING_USER]
 	                                    + sandbox->duration_of_state[SANDBOX_RUNNING_SYS];
-	execution_histogram_update(&route->execution_histogram, execution_duration);
+	execution_histogram_update(&sandbox->route->execution_histogram, execution_duration);
 #endif
 
 #ifdef ADMISSIONS_CONTROL
@@ -65,7 +60,7 @@ sandbox_set_as_complete(struct sandbox *sandbox, sandbox_state_t last_state)
 	/* Terminal State Logging for Sandbox */
 	sandbox_perf_log_print_entry(sandbox);
 	sandbox_summarize_page_allocations(sandbox);
-	route_latency_add(&route->latency, sandbox->total_time);
+	route_latency_add(&sandbox->route->latency, sandbox->total_time);
 
 	/* State Change Hooks */
 	sandbox_state_transition_from_hook(sandbox, last_state);

runtime/include/sandbox_set_as_error.h

@@ -33,40 +33,54 @@ sandbox_set_as_error(struct sandbox *sandbox, sandbox_state_t last_state)
 	uint64_t now   = __getcycles();
 
 	switch (last_state) {
-	case SANDBOX_ALLOCATED:
+	case SANDBOX_INITIALIZED:
+		assert(sandbox->memory == NULL);
 		break;
-	case SANDBOX_RUNNING_SYS: {
-		local_runqueue_delete(sandbox);
-		sandbox_free_linear_memory(sandbox);
+	case SANDBOX_PREEMPTED:
+		/* Global work-shedding scenario, where we kill a job right after pullling from global queue */
+		if (USING_LOCAL_RUNQUEUE) {
+			local_runqueue_delete(sandbox); 
+		} else {
+			assert(sandbox->owned_worker_idx == -2);
+			assert(sandbox->pq_idx_in_runqueue == 0);
+		}
+		
+		// sandbox_free_linear_memory(sandbox);
 		break;
-	}
-	default: {
+	case SANDBOX_RUNNABLE: /* When using local queues */
+	case SANDBOX_RUNNING_SYS:
+	case SANDBOX_INTERRUPTED:
+		assert(sandbox->owned_worker_idx >= 0);
+		assert(sandbox->pq_idx_in_runqueue >= 1);
+		local_runqueue_delete(sandbox); 
+		// sandbox_free_linear_memory(sandbox);
+		break;
+	default:
 		panic("Sandbox %lu | Illegal transition from %s to Error\n", sandbox->id,
 		      sandbox_state_stringify(last_state));
 	}
-	}
 
 	/* State Change Bookkeeping */
 	assert(now > sandbox->timestamp_of.last_state_change);
 	sandbox->last_state_duration = now - sandbox->timestamp_of.last_state_change;
-	if (last_state == SANDBOX_RUNNING_SYS) {
-		sandbox->remaining_exec = (sandbox->remaining_exec > sandbox->last_state_duration)
-		                            ? sandbox->remaining_exec - sandbox->last_state_duration
-		                            : 0;
-	}
+	if(last_state == SANDBOX_RUNNING_SYS) sandbox->last_running_state_duration += sandbox->last_state_duration;
 	sandbox->duration_of_state[last_state] += sandbox->last_state_duration;
 	sandbox->timestamp_of.last_state_change = now;
 	sandbox_state_history_append(&sandbox->state_history, SANDBOX_ERROR);
 	sandbox_state_totals_increment(SANDBOX_ERROR);
 	sandbox_state_totals_decrement(last_state);
 
+	sandbox->timestamp_of.completion = now;
+	sandbox->total_time              = sandbox->timestamp_of.completion - sandbox->timestamp_of.allocation;
+
 #ifdef ADMISSIONS_CONTROL
 	/* Admissions Control Post Processing */
 	admissions_control_subtract(sandbox->admissions_estimate);
 #endif
 
 	/* Return HTTP session to listener core to be written back to client */
-	http_session_set_response_header(sandbox->http, 500);
+	const int http_status_code = (sandbox->response_code > 0) ? sandbox->response_code / 10 : 500;
+	http_session_set_response_header(sandbox->http, http_status_code);
 	sandbox->http->state = HTTP_SESSION_EXECUTION_COMPLETE;
 	http_session_send_response(sandbox->http, (void_star_cb)listener_thread_register_http_session);
 	sandbox->http = NULL;
@@ -85,8 +99,8 @@ sandbox_set_as_error(struct sandbox *sandbox, sandbox_state_t last_state)
 static inline void
 sandbox_exit_error(struct sandbox *sandbox)
 {
-	assert(sandbox->state == SANDBOX_RUNNING_SYS);
-	sandbox_set_as_error(sandbox, SANDBOX_RUNNING_SYS);
+	// assert(sandbox->state == SANDBOX_RUNNING_SYS);
+	sandbox_set_as_error(sandbox, sandbox->state);
 
 	sandbox_process_scheduler_updates(sandbox);
 }

runtime/include/sandbox_set_as_initialized.h

@@ -24,14 +24,12 @@ sandbox_set_as_initialized(struct sandbox *sandbox, sandbox_state_t last_state)
 	uint64_t now   = __getcycles();
 
 	switch (last_state) {
-	case SANDBOX_ALLOCATED: {
+	case SANDBOX_ALLOCATED:
 		break;
-	}
-	default: {
+	default:
 		panic("Sandbox %lu | Illegal transition from %s to Preempted\n", sandbox->id,
 		      sandbox_state_stringify(last_state));
 	}
-	}
 
 	/* State Change Bookkeeping */
 	assert(now > sandbox->timestamp_of.last_state_change);

runtime/include/sandbox_set_as_interrupted.h

@@ -15,6 +15,7 @@ static inline void
 sandbox_set_as_interrupted(struct sandbox *sandbox, sandbox_state_t last_state)
 {
 	assert(sandbox);
+	assert(last_state == SANDBOX_RUNNING_USER);
 
 	/* WARNING: All code before this assignment is preemptable */
 	sandbox->state = SANDBOX_INTERRUPTED;
@@ -25,10 +26,7 @@ sandbox_set_as_interrupted(struct sandbox *sandbox, sandbox_state_t last_state)
 	/* State Change Bookkeeping */
 	assert(now > sandbox->timestamp_of.last_state_change);
 	sandbox->last_state_duration = now - sandbox->timestamp_of.last_state_change;
-	assert(last_state == SANDBOX_RUNNING_USER);
-	sandbox->remaining_exec = (sandbox->remaining_exec > sandbox->last_state_duration)
-	                            ? sandbox->remaining_exec - sandbox->last_state_duration
-	                            : 0;
+	sandbox->last_running_state_duration += sandbox->last_state_duration;
 	sandbox->duration_of_state[last_state] += sandbox->last_state_duration;
 	sandbox->timestamp_of.last_state_change = now;
 	/* We do not append SANDBOX_INTERRUPTED to the sandbox_state_history because it would quickly fill the buffer */
@@ -44,8 +42,6 @@ static inline void
 sandbox_interrupt(struct sandbox *sandbox)
 {
 	sandbox_set_as_interrupted(sandbox, sandbox->state);
-
-	sandbox_process_scheduler_updates(sandbox);
 }
 
 
@@ -69,9 +65,9 @@ sandbox_interrupt_return(struct sandbox *sandbox, sandbox_state_t interrupted_st
 	sandbox_state_totals_increment(interrupted_state);
 	sandbox_state_totals_decrement(SANDBOX_INTERRUPTED);
 
-	if (sandbox->absolute_deadline < now) {
-		// printf("Interrupted Sandbox missed deadline already!\n");
-	}
+	// if (sandbox->absolute_deadline < now) {
+	// 	printf("Interrupted sandbox #%lu of %s missed deadline in worker #%d!\n", sandbox->id, sandbox->tenant->name, worker_thread_idx);
+	// }
 
 	barrier();
 	/* WARNING: Code after this assignment may be preemptable */

runtime/include/sandbox_set_as_preempted.h

@@ -27,14 +27,12 @@ sandbox_set_as_preempted(struct sandbox *sandbox, sandbox_state_t last_state)
 	uint64_t now   = __getcycles();
 
 	switch (last_state) {
-	case SANDBOX_INTERRUPTED: {
+	case SANDBOX_INTERRUPTED:
 		break;
-	}
-	default: {
+	default: 
 		panic("Sandbox %lu | Illegal transition from %s to Preempted\n", sandbox->id,
 		      sandbox_state_stringify(last_state));
 	}
-	}
 
 	/* State Change Bookkeeping */
 	assert(now > sandbox->timestamp_of.last_state_change);
@@ -55,4 +53,6 @@ sandbox_preempt(struct sandbox *sandbox)
 {
 	assert(sandbox->state == SANDBOX_INTERRUPTED);
 	sandbox_set_as_preempted(sandbox, SANDBOX_INTERRUPTED);
+
+	// sandbox_process_scheduler_updates(sandbox);
 }

runtime/include/sandbox_set_as_returned.h

@@ -31,29 +31,28 @@ sandbox_set_as_returned(struct sandbox *sandbox, sandbox_state_t last_state)
 	uint64_t now   = __getcycles();
 
 	switch (last_state) {
-	case SANDBOX_RUNNING_SYS: {
+	case SANDBOX_RUNNING_SYS:
 		local_runqueue_delete(sandbox);
-		sandbox_free_linear_memory(sandbox);
+		// sandbox_free_linear_memory(sandbox);
 		break;
-	}
-	default: {
+	default:
 		panic("Sandbox %lu | Illegal transition from %s to Returned\n", sandbox->id,
 		      sandbox_state_stringify(last_state));
 	}
-	}
 
 	/* State Change Bookkeeping */
 	assert(now > sandbox->timestamp_of.last_state_change);
 	sandbox->last_state_duration = now - sandbox->timestamp_of.last_state_change;
-	sandbox->remaining_exec      = (sandbox->remaining_exec > sandbox->last_state_duration)
-	                                 ? sandbox->remaining_exec - sandbox->last_state_duration
-	                                 : 0;
+	sandbox->last_running_state_duration += sandbox->last_state_duration;
 	sandbox->duration_of_state[last_state] += sandbox->last_state_duration;
 	sandbox->timestamp_of.last_state_change = now;
 	sandbox_state_history_append(&sandbox->state_history, SANDBOX_RETURNED);
 	sandbox_state_totals_increment(SANDBOX_RETURNED);
 	sandbox_state_totals_decrement(last_state);
 
+	sandbox->timestamp_of.completion = now;
+	sandbox->total_time = sandbox->timestamp_of.completion - sandbox->timestamp_of.allocation;
+
 	http_session_set_response_header(sandbox->http, 200);
 	sandbox->http->state = HTTP_SESSION_EXECUTION_COMPLETE;
 	http_session_send_response(sandbox->http, (void_star_cb)listener_thread_register_http_session);

runtime/include/sandbox_set_as_runnable.h

@@ -29,20 +29,20 @@ sandbox_set_as_runnable(struct sandbox *sandbox, sandbox_state_t last_state)
 	uint64_t now   = __getcycles();
 
 	switch (last_state) {
-	case SANDBOX_INITIALIZED: {
-		sandbox->timestamp_of.dispatched = now;
+	case SANDBOX_INITIALIZED: 
+		if(sandbox->timestamp_of.dispatched == 0) sandbox->timestamp_of.dispatched = now;
 		local_runqueue_add(sandbox);
+		// sandbox->owned_worker_idx = worker_thread_idx;
+		assert(sandbox->original_owner_worker_idx == -2);
+		sandbox->original_owner_worker_idx = worker_thread_idx;
 		break;
-	}
-	case SANDBOX_ASLEEP: {
+	case SANDBOX_ASLEEP:
 		local_runqueue_add(sandbox);
 		break;
-	}
-	default: {
+	default: 
 		panic("Sandbox %lu | Illegal transition from %s to Runnable\n", sandbox->id,
 		      sandbox_state_stringify(last_state));
 	}
-	}
 
 	/* State Change Bookkeeping */
 	assert(now > sandbox->timestamp_of.last_state_change);
@@ -63,5 +63,7 @@ static inline void
 sandbox_wakeup(struct sandbox *sandbox)
 {
 	assert(sandbox->state == SANDBOX_ASLEEP);
+	assert(0); // Now sandbox should not sleep
+
 	sandbox_set_as_runnable(sandbox, SANDBOX_ASLEEP);
 }

runtime/include/sandbox_set_as_running_sys.h

@@ -23,29 +23,22 @@ sandbox_set_as_running_sys(struct sandbox *sandbox, sandbox_state_t last_state)
 	uint64_t now = __getcycles();
 
 	switch (last_state) {
-	case SANDBOX_RUNNING_USER: {
+	case SANDBOX_RUNNING_USER: 
 		assert(sandbox == current_sandbox_get());
 		assert(runtime_worker_threads_deadline[worker_thread_idx] == sandbox->absolute_deadline);
 		break;
-	}
-	case SANDBOX_RUNNABLE: {
+	case SANDBOX_RUNNABLE: 
 		assert(sandbox);
 		break;
-	}
-	default: {
+	default: 
 		panic("Sandbox %lu | Illegal transition from %s to Running Sys\n", sandbox->id,
 		      sandbox_state_stringify(last_state));
 	}
-	}
 
 	/* State Change Bookkeeping */
 	assert(now > sandbox->timestamp_of.last_state_change);
 	sandbox->last_state_duration = now - sandbox->timestamp_of.last_state_change;
-	if (last_state == SANDBOX_RUNNING_USER) {
-		sandbox->remaining_exec = (sandbox->remaining_exec > sandbox->last_state_duration)
-		                            ? sandbox->remaining_exec - sandbox->last_state_duration
-		                            : 0;
-	}
+	if(last_state == SANDBOX_RUNNING_USER) sandbox->last_running_state_duration += sandbox->last_state_duration;
 	sandbox->duration_of_state[last_state] += sandbox->last_state_duration;
 	sandbox->timestamp_of.last_state_change = now;
 	sandbox_state_history_append(&sandbox->state_history, SANDBOX_RUNNING_SYS);
@@ -64,6 +57,4 @@ sandbox_syscall(struct sandbox *sandbox)
 
 	assert(sandbox->state == SANDBOX_RUNNING_USER);
 	sandbox_set_as_running_sys(sandbox, SANDBOX_RUNNING_USER);
-
-	sandbox_process_scheduler_updates(sandbox);
 }

runtime/include/sandbox_set_as_running_user.h

@@ -19,29 +19,25 @@ sandbox_set_as_running_user(struct sandbox *sandbox, sandbox_state_t last_state)
 	uint64_t now = __getcycles();
 
 	switch (last_state) {
-	case SANDBOX_RUNNING_SYS: {
+	case SANDBOX_RUNNING_SYS: 
 		assert(sandbox == current_sandbox_get());
 		assert(runtime_worker_threads_deadline[worker_thread_idx] == sandbox->absolute_deadline);
 		break;
-	}
-	case SANDBOX_PREEMPTED: {
+	
+	case SANDBOX_PREEMPTED: 
 		break;
-	}
-	default: {
+	
+	default: 
 		panic("Sandbox %lu | Illegal transition from %s to Running\n", sandbox->id,
 		      sandbox_state_stringify(last_state));
-	}
+	
 	}
 
 
 	/* State Change Bookkeeping */
 	assert(now > sandbox->timestamp_of.last_state_change);
 	sandbox->last_state_duration = now - sandbox->timestamp_of.last_state_change;
-	if (last_state == SANDBOX_RUNNING_SYS) {
-		sandbox->remaining_exec = (sandbox->remaining_exec > sandbox->last_state_duration)
-		                            ? sandbox->remaining_exec - sandbox->last_state_duration
-		                            : 0;
-	}
+	if(last_state == SANDBOX_RUNNING_SYS) sandbox->last_running_state_duration += sandbox->last_state_duration;
 	sandbox->duration_of_state[last_state] += sandbox->last_state_duration;
 	sandbox->timestamp_of.last_state_change = now;
 	sandbox_state_history_append(&sandbox->state_history, SANDBOX_RUNNING_USER);
@@ -52,8 +48,7 @@ sandbox_set_as_running_user(struct sandbox *sandbox, sandbox_state_t last_state)
 	sandbox_state_transition_from_hook(sandbox, last_state);
 	sandbox_state_transition_to_hook(sandbox, SANDBOX_RUNNING_USER);
 
-	if (last_state == SANDBOX_RUNNING_SYS)
-		sandbox_process_scheduler_updates(sandbox); // TODO: is this code preemptable? Ok to be?
+	assert(sandbox->memory->abi.size > 0);
 
 	barrier();
 	sandbox->state = SANDBOX_RUNNING_USER;

runtime/include/sandbox_types.h

@@ -15,6 +15,11 @@
 #include "wasm_stack.h"
 #include "wasm_types.h"
 
+// #include "wasm_globals.h"
+// #include "wasi.h"
+// #include "listener_thread.h"
+// #include "ck_ring.h"
+
 /*********************
  * Structs and Types *
  ********************/
@@ -30,6 +35,7 @@ struct sandbox_timestamps {
 #endif
 };
 
+struct sandbox_metadata;
 struct sandbox {
 	/* used by ps_list's default name-based MACROS for the scheduling runqueue */
 	/* Keep as first member of sandbox struct to ensure ps_list maintains alignment */
@@ -39,6 +45,14 @@ struct sandbox {
 	sandbox_state_t              state;
 	struct sandbox_state_history state_history;
 	uint16_t                     response_code;
+	
+	size_t                       pq_idx_in_runqueue;
+	size_t                       pq_idx_in_tenant_queue;
+	int                          owned_worker_idx;
+	int                          original_owner_worker_idx;
+	int                          global_queue_type;
+	uint8_t                      num_of_overshoots;
+	struct sandbox_metadata     *sandbox_meta;
 
 
 	/* Accounting Info */
@@ -57,13 +71,20 @@ struct sandbox {
 	struct wasm_memory      *memory;
 	struct vec_wasm_global_t globals;
 
+	// uint64_t sizes[1000000];
+	// uint64_t sizesize;
+
 	/* Scheduling and Temporal State */
 	struct sandbox_timestamps timestamp_of;
 	uint64_t                  duration_of_state[SANDBOX_STATE_COUNT];
 	uint64_t                  last_state_duration;
+	uint64_t                  last_running_state_duration;
 
 	uint64_t remaining_exec;
 	uint64_t absolute_deadline;
+	bool     exceeded_estimation;
+	bool     writeback_preemption_in_progress;
+	bool     writeback_overshoot_in_progress;
 	uint64_t admissions_estimate; /* estimated execution time (cycles) * runtime_admissions_granularity / relative
 	                                 deadline (cycles) */
 	uint64_t total_time;          /* Total time from Request to Response */
@@ -72,5 +93,29 @@ struct sandbox {
 	/* System Interface State */
 	int32_t         return_value;
 	wasi_context_t *wasi_context;
-
 } PAGE_ALIGNED;
+
+struct sandbox_metadata {
+	struct sandbox        *sandbox_shadow;
+	struct tenant         *tenant;
+	struct route          *route;
+	struct priority_queue *tenant_queue;
+	uint64_t               id;
+	uint64_t               allocation_timestamp;
+	uint64_t               absolute_deadline;
+	uint64_t               remaining_exec;
+	uint64_t               total_be_exec_cycles;
+	uint64_t               extra_slack; /* cycles */
+	size_t                 pq_idx_in_tenant_queue;
+	int                    owned_worker_idx;
+	sandbox_state_t        state;
+	bool                   exceeded_estimation;
+	bool                   terminated;
+	int                    global_queue_type;
+	int                    worker_id_virt;
+	uint16_t               error_code;
+
+	struct job_node    *trs_job_node;
+	struct demand_node *demand_node;
+	struct demand_node *local_dbf_demand_node;
+}; // PAGE_ALIGNED;

runtime/include/scheduler.h

@@ -9,11 +9,13 @@
 #include "global_request_scheduler_deque.h"
 #include "global_request_scheduler_minheap.h"
 #include "global_request_scheduler_mtds.h"
+#include "global_request_scheduler_mtdbf.h"
 #include "local_cleanup_queue.h"
 #include "local_runqueue.h"
 #include "local_runqueue_list.h"
 #include "local_runqueue_minheap.h"
 #include "local_runqueue_mtds.h"
+#include "local_runqueue_mtdbf.h"
 #include "panic.h"
 #include "sandbox_functions.h"
 #include "sandbox_set_as_interrupted.h"
@@ -22,6 +24,7 @@
 #include "sandbox_set_as_running_sys.h"
 #include "sandbox_set_as_running_user.h"
 #include "sandbox_types.h"
+#include "sandbox_set_as_error.h"
 #include "scheduler_options.h"
 
 
@@ -66,7 +69,64 @@
 static inline struct sandbox *
 scheduler_mtdbf_get_next()
 {
-	return NULL;
+	/* Get the deadline of the sandbox at the head of the local queue */
+	struct sandbox *local          = local_runqueue_get_next();
+	uint64_t        local_deadline = local == NULL ? UINT64_MAX : local->absolute_deadline;
+
+	uint64_t local_rem = local == NULL ? 0 : local->remaining_exec;
+	struct sandbox *global         = NULL;
+	uint64_t now = __getcycles();
+
+	struct sandbox_metadata global_metadata = global_request_scheduler_peek_metadata();
+
+	/* Try to pull and allocate from the global queue if earlier
+	 * This will be placed at the head of the local runqueue */
+	if(USING_EARLIEST_START_FIRST) {
+		if (global_metadata.absolute_deadline - global_metadata.remaining_exec >= local_deadline - local_rem) goto done;
+	} else {
+		if (global_metadata.absolute_deadline >= local_deadline) goto done;
+	}
+
+	if (global_request_scheduler_remove_if_earlier(&global, local_deadline) == 0) {
+		assert(global != NULL);
+		// assert(global->absolute_deadline < local_deadline);
+		if (sandbox_validate_self_lifetime(global) == 0) {
+			if (global->state == SANDBOX_INITIALIZED) {
+				sandbox_prepare_execution_environment(global);
+				sandbox_set_as_runnable(global, SANDBOX_INITIALIZED);
+				
+				struct comm_with_worker *cfw = &comm_from_workers[worker_thread_idx];
+				assert(cfw);
+
+				struct message new_message = {
+					.sandbox                     = global,
+					.sandbox_id                  = global->id,
+					.sandbox_meta                = global->sandbox_meta,
+					.state                       = global->state,
+					.sender_worker_idx           = worker_thread_idx,
+					.exceeded_estimation         = global->exceeded_estimation,
+					.message_type                = MESSAGE_CFW_PULLED_NEW_SANDBOX,
+					.timestamp = now
+				};
+
+				if (!ck_ring_enqueue_spsc_message(&cfw->worker_ring, cfw->worker_ring_buffer, &new_message)) {
+					panic("Ring The buffer was full and the enqueue operation has failed.!");
+				}
+			} else {
+				assert(global->state == SANDBOX_PREEMPTED);
+				// debuglog("Resuming writeback\n");
+				local_runqueue_add(global);
+				// global->owned_worker_idx = worker_thread_idx;
+			}
+
+			assert(global->state == SANDBOX_RUNNABLE || global->state == SANDBOX_PREEMPTED);
+			// printf("Worker %i accepted a sandbox #%lu!\n", worker_thread_idx, global->id);
+		}
+	}
+
+done:
+	/* Return what is at the head of the local runqueue or NULL if empty */
+	return local_runqueue_get_next();
 }
 
 static inline struct sandbox *
@@ -204,7 +264,7 @@ scheduler_initialize()
 {
 	switch (scheduler) {
 	case SCHEDULER_MTDBF:
-		/* TODO: loading */
+		global_request_scheduler_mtdbf_initialize();
 		break;
 	case SCHEDULER_MTDS:
 		global_request_scheduler_mtds_initialize();
@@ -226,7 +286,7 @@ scheduler_runqueue_initialize()
 {
 	switch (scheduler) {
 	case SCHEDULER_MTDBF:
-		// local_runqueue_mtdbf_initialize();
+		local_runqueue_mtdbf_initialize();
 		break;
 	case SCHEDULER_MTDS:
 		local_runqueue_mtds_initialize();
@@ -284,12 +344,20 @@ scheduler_log_sandbox_switch(struct sandbox *current_sandbox, struct sandbox *ne
 static inline void
 scheduler_preemptive_switch_to(ucontext_t *interrupted_context, struct sandbox *next)
 {
+	/* Switch to base context */
+	if (next == NULL) {
+		arch_context_restore_fast(&interrupted_context->uc_mcontext, &worker_thread_base_context);
+		current_sandbox_set(NULL);
+		return;
+	}
+
 	/* Switch to next sandbox */
 	switch (next->ctxt.variant) {
 	case ARCH_CONTEXT_VARIANT_FAST: {
 		assert(next->state == SANDBOX_RUNNABLE);
 		arch_context_restore_fast(&interrupted_context->uc_mcontext, &next->ctxt);
 		current_sandbox_set(next);
+assert(sledge_abi__current_wasm_module_instance.abi.memory.id == next->id);
 		sandbox_set_as_running_sys(next, SANDBOX_RUNNABLE);
 		break;
 	}
@@ -297,6 +365,7 @@ scheduler_preemptive_switch_to(ucontext_t *interrupted_context, struct sandbox *
 		assert(next->state == SANDBOX_PREEMPTED);
 		arch_context_restore_slow(&interrupted_context->uc_mcontext, &next->ctxt);
 		current_sandbox_set(next);
+assert(sledge_abi__current_wasm_module_instance.abi.memory.id == next->id);
 		sandbox_set_as_running_user(next, SANDBOX_PREEMPTED);
 		break;
 	}
@@ -307,6 +376,53 @@ scheduler_preemptive_switch_to(ucontext_t *interrupted_context, struct sandbox *
 	}
 }
 
+static inline int
+scheduler_check_messages_from_listener()
+{
+	int rc = 0;
+
+	assert(comm_to_workers);
+
+	struct message           new_message = { 0 };
+	struct comm_with_worker *ctw         = &comm_to_workers[worker_thread_idx];
+	assert(ctw);
+	assert(ctw->worker_idx == worker_thread_idx);
+	assert(ck_ring_size(&ctw->worker_ring) < LISTENER_THREAD_RING_SIZE);
+
+	while (ck_ring_dequeue_spsc_message(&ctw->worker_ring, ctw->worker_ring_buffer, &new_message)) {
+		assert(new_message.message_type == MESSAGE_CTW_SHED_CURRENT_JOB);
+		/* Check if the sandbox is still alive (not freed yet) */
+		if (sandbox_refs[new_message.sandbox_id % RUNTIME_MAX_ALIVE_SANDBOXES]) {
+			struct sandbox *sandbox_to_kill = new_message.sandbox;
+			assert(sandbox_to_kill);
+			assert(sandbox_to_kill->id == new_message.sandbox_id);
+
+			if (sandbox_to_kill->pq_idx_in_runqueue == 0 || sandbox_to_kill->owned_worker_idx != worker_thread_idx) {
+			/* Make sure the sandbox is in a non-terminal or asleep state (aka: still in the runqueue) */
+				new_message.sandbox = NULL;
+				new_message.sandbox_id = 0;
+				continue;
+			}
+
+			struct sandbox_metadata *sandbox_meta = sandbox_to_kill->sandbox_meta;
+			assert(sandbox_meta);
+			assert(sandbox_meta->sandbox_shadow == sandbox_to_kill);
+			assert(sandbox_meta->id == sandbox_to_kill->id);
+			assert(sandbox_meta->error_code > 0);
+
+			// printf("Worker#%d shedding sandbox #%lu\n", worker_thread_idx, sandbox_to_kill->id);
+			assert(sandbox_to_kill->response_code == 0);
+			sandbox_to_kill->response_code = sandbox_meta->error_code;
+			sandbox_exit_error(sandbox_to_kill);
+			local_cleanup_queue_add(sandbox_to_kill);
+		}
+		new_message.sandbox = NULL;
+		new_message.sandbox_id = 0;
+	}
+
+	return rc;
+}
+
 /**
  * Call either at preemptions or blockings to update the scheduler-specific
  *  properties for the given tenant.
@@ -316,15 +432,16 @@ scheduler_process_policy_specific_updates_on_interrupts(struct sandbox *interrup
 {
 	switch (scheduler) {
 	case SCHEDULER_FIFO:
-		return;
 	case SCHEDULER_EDF:
 	case SCHEDULER_SJF:
+		sandbox_process_scheduler_updates(interrupted_sandbox);
 		return;
 	case SCHEDULER_MTDS:
+		sandbox_process_scheduler_updates(interrupted_sandbox);
 		local_timeout_queue_process_promotions();
 		return;
 	case SCHEDULER_MTDBF:
-		// scheduler_check_messages_from_listener();
+		scheduler_check_messages_from_listener();
 		if (interrupted_sandbox->state != SANDBOX_ERROR) {
 			sandbox_process_scheduler_updates(interrupted_sandbox);
 		}
@@ -343,17 +460,25 @@ scheduler_preemptive_sched(ucontext_t *interrupted_context)
 {
 	assert(interrupted_context != NULL);
 
-	/* Process epoll to make sure that all runnable jobs are considered for execution */
-
 	struct sandbox *interrupted_sandbox = current_sandbox_get();
 	assert(interrupted_sandbox != NULL);
 	assert(interrupted_sandbox->state == SANDBOX_INTERRUPTED);
-	// printf ("Worker #%d interrupted sandbox #%lu\n", worker_thread_idx, interrupted_sandbox->id);
+
 	scheduler_process_policy_specific_updates_on_interrupts(interrupted_sandbox);
 
 	struct sandbox *next = scheduler_get_next();
+
+	/* Assumption: the current sandbox is still there, even if the worker had to shed it from its runqueue above */
+	assert(interrupted_sandbox != NULL);
+
+	if (interrupted_sandbox->state == SANDBOX_ERROR) goto done;
+	if(!(interrupted_sandbox->state == SANDBOX_INTERRUPTED)) {
+		printf("sand state: %u\n", interrupted_sandbox->state);
+	}
+	assert(interrupted_sandbox->state == SANDBOX_INTERRUPTED);
+
 	/* Assumption: the current sandbox is on the runqueue, so the scheduler should always return something */
-	assert(next != NULL);
+	// assert(next != NULL); // Cannot assert, since the head of the global queue may have expired and cleaned before this
 
 	/* If current equals next, no switch is necessary, so resume execution */
 	if (interrupted_sandbox == next) {
@@ -369,11 +494,48 @@ scheduler_preemptive_sched(ucontext_t *interrupted_context)
 	scheduler_log_sandbox_switch(interrupted_sandbox, next);
 	sandbox_preempt(interrupted_sandbox);
 
-	// Write back global at idx 0
-	wasm_globals_set_i64(&interrupted_sandbox->globals, 0, sledge_abi__current_wasm_module_instance.abi.wasmg_0,
-	                     true);
-
+	// Update global at idx 0
+	int rc = wasm_globals_set_i64(&interrupted_sandbox->globals, 0,
+							sledge_abi__current_wasm_module_instance.abi.wasmg_0, true);
+	assert(rc == 0);
+	
 	arch_context_save_slow(&interrupted_sandbox->ctxt, &interrupted_context->uc_mcontext);
+
+#ifdef TRAFFIC_CONTROL
+	if (USING_WRITEBACK_FOR_PREEMPTION || USING_WRITEBACK_FOR_OVERSHOOT) {
+		struct message new_message = {
+			.sandbox             = interrupted_sandbox,
+			.sandbox_id          = interrupted_sandbox->id,
+			.sandbox_meta        = interrupted_sandbox->sandbox_meta,
+			.state               = interrupted_sandbox->state,
+			.sender_worker_idx   = worker_thread_idx,
+			.exceeded_estimation = interrupted_sandbox->exceeded_estimation,
+			.timestamp           = interrupted_sandbox->timestamp_of.last_state_change,
+			.remaining_exec = interrupted_sandbox->remaining_exec
+		};
+
+		if (interrupted_sandbox->writeback_overshoot_in_progress) {
+			assert(USING_WRITEBACK_FOR_OVERSHOOT);
+			assert(interrupted_sandbox->remaining_exec == 0);
+			new_message.message_type = MESSAGE_CFW_WRITEBACK_OVERSHOOT;
+			new_message.adjustment = runtime_quantum;
+		}
+		else if (interrupted_sandbox->writeback_preemption_in_progress) {
+			assert(USING_WRITEBACK_FOR_PREEMPTION);
+			assert(USING_LOCAL_RUNQUEUE == false);
+			new_message.message_type = MESSAGE_CFW_WRITEBACK_PREEMPTION;
+			new_message.adjustment = 0;
+		} else panic("No writeback is in progress. Cannot be here!");
+
+		struct comm_with_worker *cfw = &comm_from_workers[worker_thread_idx];
+		if (!ck_ring_enqueue_spsc_message(&cfw->worker_ring, cfw->worker_ring_buffer, &new_message)) {
+			panic("Ring The buffer was full and the enqueue operation has failed.!");
+		}
+	}
+#endif
+
+	/* CAUTION! Worker MUST NOT access interrupted sandbox after this point! */
+done:
 	scheduler_preemptive_switch_to(interrupted_context, next);
 }
 
@@ -394,12 +556,14 @@ scheduler_cooperative_switch_to(struct arch_context *current_context, struct san
 	case SANDBOX_RUNNABLE: {
 		assert(next_context->variant == ARCH_CONTEXT_VARIANT_FAST);
 		current_sandbox_set(next_sandbox);
+assert(sledge_abi__current_wasm_module_instance.abi.memory.id == next_sandbox->id);
 		sandbox_set_as_running_sys(next_sandbox, SANDBOX_RUNNABLE);
 		break;
 	}
 	case SANDBOX_PREEMPTED: {
 		assert(next_context->variant == ARCH_CONTEXT_VARIANT_SLOW);
 		current_sandbox_set(next_sandbox);
+assert(sledge_abi__current_wasm_module_instance.abi.memory.id == next_sandbox->id);
 		/* arch_context_switch triggers a SIGUSR1, which transitions next_sandbox to running_user */
 		break;
 	}
@@ -424,6 +588,7 @@ scheduler_switch_to_base_context(struct arch_context *current_context)
 static inline void
 scheduler_idle_loop()
 {
+	int spin = 0, max_spin = 0;
 	while (true) {
 		/* Assumption: only called by the "base context" */
 		assert(current_sandbox_get() == NULL);
@@ -435,6 +600,13 @@ scheduler_idle_loop()
 		struct sandbox *next_sandbox = scheduler_get_next();
 		if (next_sandbox != NULL) {
 			scheduler_cooperative_switch_to(&worker_thread_base_context, next_sandbox);
+			spin++;
+			if (spin > max_spin) {
+				max_spin = spin;
+				// printf("Worker #%d max useless spins #%d!\n", worker_thread_idx, max_spin);
+			}
+		} else {
+			spin = 0;
 		}
 
 		/* Clear the cleanup queue */
@@ -479,6 +651,7 @@ scheduler_cooperative_sched(bool add_to_cleanup_queue)
 
 	/* If our sandbox slept and immediately woke up, we can just return */
 	if (next_sandbox == exiting_sandbox) {
+		assert(0); // Never happens, sandboxes don't sleep anymore
 		sandbox_set_as_running_sys(next_sandbox, SANDBOX_RUNNABLE);
 		current_sandbox_set(next_sandbox);
 		return;
@@ -487,6 +660,7 @@ scheduler_cooperative_sched(bool add_to_cleanup_queue)
 	scheduler_log_sandbox_switch(exiting_sandbox, next_sandbox);
 
 	// Write back global at idx 0
+	assert(sledge_abi__current_wasm_module_instance.abi.wasmg_0 == 0);
 	wasm_globals_set_i64(&exiting_sandbox->globals, 0, sledge_abi__current_wasm_module_instance.abi.wasmg_0, true);
 
 	if (add_to_cleanup_queue) local_cleanup_queue_add(exiting_sandbox);
@@ -503,8 +677,11 @@ scheduler_cooperative_sched(bool add_to_cleanup_queue)
 static inline bool
 scheduler_worker_would_preempt(int worker_idx)
 {
-	assert(scheduler == SCHEDULER_EDF);
+	// assert(scheduler == SCHEDULER_EDF);
 	uint64_t local_deadline  = runtime_worker_threads_deadline[worker_idx];
 	uint64_t global_deadline = global_request_scheduler_peek();
-	return global_deadline < local_deadline;
+	
+	/* Only send a worker SIGARLM if it has a sandbox to execute (MTDBF) 
+	    or it needs to check the global queue for a new higher priority job */
+	return local_deadline < UINT64_MAX || global_deadline < local_deadline;
 }

runtime/include/sledge_abi_symbols.h

@@ -9,7 +9,7 @@
 #include "wasm_types.h"
 
 struct sledge_abi_symbols {
-	void                         *handle;
+	void                         *handle; //////////////// TODO: Maybe make this local below?
 	sledge_abi__init_globals_fn_t initialize_globals;
 	sledge_abi__init_mem_fn_t     initialize_memory;
 	sledge_abi__init_tbl_fn_t     initialize_tables;

runtime/include/tenant.h

@@ -13,7 +13,8 @@ enum MULTI_TENANCY_CLASS
 };
 
 struct tenant_timeout {
-	uint64_t                               timeout;
+	uint64_t timeout;
+
 	struct tenant                         *tenant;
 	struct perworker_tenant_sandbox_queue *pwt;
 };
@@ -32,6 +33,21 @@ struct tenant_global_request_queue {
 	_Atomic volatile enum MULTI_TENANCY_CLASS mt_class;
 };
 
+struct job_node {
+	struct ps_list           list;
+	uint64_t                 exec;
+	uint64_t                 timestamp;
+	struct sandbox_metadata *sandbox_meta;
+};
+
+struct tenant_reservation_server {
+	uint64_t            max_budget_guaranteed;
+	uint64_t            budget_guaranteed;
+	uint64_t            budget_best;
+	struct ps_list_head admitted_jobs_list;
+	struct ps_list_head admitted_BE_jobs_list;
+};
+
 struct tenant {
 	enum epoll_tag         tag; /* Tag must be first member */
 	char                  *name;
@@ -40,13 +56,19 @@ struct tenant {
 	struct module_database module_db;
 	struct map             scratch_storage;
 
-	/* Deferrable Server Attributes */
-	uint64_t                 replenishment_period; /* cycles, not changing after init */
-	uint64_t                 max_budget;           /* cycles, not changing after init */
-	_Atomic volatile int64_t remaining_budget;     /* cycles left till next replenishment, can be negative */
+	/* Multi-Tenancy Attributes */
+	uint64_t                 max_relative_deadline;
+	uint64_t                 replenishment_period;   /* cycles, not changing after init */
+	uint64_t                 max_budget;             /* cycles, not changing after init */
+	_Atomic volatile int64_t remaining_budget;       /* cycles left till next replenishment, can be negative */
+	uint8_t                  reservation_percentile; /* percentile of the overall reservation utilisation */
 
 	struct perworker_tenant_sandbox_queue *pwt_sandboxes;
 	struct tenant_global_request_queue    *tgrq_requests;
+	struct priority_queue                 *local_sandbox_metas, *global_sandbox_metas;
+	struct tenant_reservation_server       trs;
+	uint32_t                               num_of_overshooted_sandboxes;
+	uint8_t                                max_overshoot_of_same_sandbox;
 };
 
 

runtime/include/tenant_config.h

@@ -14,6 +14,7 @@ enum tenant_config_member
 	tenant_config_member_port,
 	tenant_config_member_replenishment_period_us,
 	tenant_config_member_max_budget_us,
+	tenant_config_reservation_percentile,
 	tenant_config_member_routes,
 	tenant_config_member_len
 };
@@ -23,8 +24,10 @@ struct tenant_config {
 	uint16_t             port;
 	uint32_t             replenishment_period_us;
 	uint32_t             max_budget_us;
+	uint8_t              reservation_percentile;
 	struct route_config *routes;
 	size_t               routes_len;
+	uint32_t             max_relative_deadline_us;
 };
 
 static inline void
@@ -34,6 +37,7 @@ tenant_config_deinit(struct tenant_config *config)
 	config->name                    = NULL;
 	config->replenishment_period_us = 0;
 	config->max_budget_us           = 0;
+	config->reservation_percentile  = 0;
 	for (int i = 0; i < config->routes_len; i++) { route_config_deinit(&config->routes[i]); }
 	free(config->routes);
 	config->routes     = NULL;
@@ -48,9 +52,12 @@ tenant_config_print(struct tenant_config *config)
 	if (scheduler == SCHEDULER_MTDS) {
 		printf("[Tenant] Replenishment Period (us): %u\n", config->replenishment_period_us);
 		printf("[Tenant] Max Budget (us): %u\n", config->max_budget_us);
+	} else if (scheduler == SCHEDULER_MTDBF) {
+		printf("[Tenant] Reservation (%%): %u\n", config->reservation_percentile);
 	}
 	printf("[Tenant] Routes Size: %zu\n", config->routes_len);
 	for (int i = 0; i < config->routes_len; i++) { route_config_print(&config->routes[i]); }
+	printf("\n");
 }
 
 static inline int
@@ -93,6 +100,17 @@ tenant_config_validate(struct tenant_config *config, bool *did_set)
 			        (uint32_t)RUNTIME_RELATIVE_DEADLINE_US_MAX, config->max_budget_us);
 			return -1;
 		}
+	} else if (scheduler == SCHEDULER_MTDBF) {
+		if (did_set[tenant_config_reservation_percentile] == false) {
+			fprintf(stderr, "reservation-percentile field is required\n");
+			return -1;
+		}
+
+		if (config->replenishment_period_us > 100) {
+			fprintf(stderr, "Reservation-percentile must be between 0 and 100, was %u\n",
+			        config->reservation_percentile);
+			return -1;
+		}
 	}
 
 	if (config->routes_len == 0) {

runtime/include/tenant_config_parse.h

@@ -9,8 +9,9 @@
 #include "route_config_parse.h"
 #include "tenant_config.h"
 
-static const char *tenant_config_json_keys[tenant_config_member_len] = {"name", "port", "replenishment-period-us",
-                                                                        "max-budget-us", "routes"};
+static const char *tenant_config_json_keys[tenant_config_member_len] = {
+	"name", "port", "replenishment-period-us", "max-budget-us", "reservation-percentile", "routes"
+};
 
 static inline int
 tenant_config_set_key_once(bool *did_set, enum tenant_config_member member)
@@ -79,6 +80,14 @@ tenant_config_parse(struct tenant_config *config, const char *json_buf, jsmntok_
 			                        tenant_config_json_keys[tenant_config_member_max_budget_us],
 			                        &config->max_budget_us);
 			if (rc < 0) return -1;
+		} else if (strcmp(key, tenant_config_json_keys[tenant_config_reservation_percentile]) == 0) {
+			if (!has_valid_type(tokens[i], key, JSMN_PRIMITIVE, json_buf)) return -1;
+			if (tenant_config_set_key_once(did_set, tenant_config_reservation_percentile) == -1) return -1;
+
+			int rc = parse_uint8_t(tokens[i], json_buf,
+			                       tenant_config_json_keys[tenant_config_reservation_percentile],
+			                       &config->reservation_percentile);
+			if (rc < 0) return -1;
 		} else if (strcmp(key, tenant_config_json_keys[tenant_config_member_routes]) == 0) {
 			if (!has_valid_type(tokens[i], key, JSMN_ARRAY, json_buf)) return -1;
 			if (tenant_config_set_key_once(did_set, tenant_config_member_routes) == -1) return -1;
@@ -92,6 +101,9 @@ tenant_config_parse(struct tenant_config *config, const char *json_buf, jsmntok_
 				i++;
 				i = route_config_parse(&(config->routes)[route_idx], json_buf, tokens, i, tokens_size);
 				if (i == -1) return -1;
+				if (config->routes[route_idx].relative_deadline_us > config->max_relative_deadline_us) {
+					config->max_relative_deadline_us = config->routes[route_idx].relative_deadline_us;
+				}
 			}
 
 		} else {

runtime/include/tenant_functions.h

@@ -12,17 +12,37 @@
 #include "scheduler_options.h"
 #include "tenant.h"
 #include "tenant_config.h"
+#include "priority_queue.h"
+#include "sandbox_functions.h"
+#include "dbf.h"
+
+#define REPLENISHMENT_PERIOD (runtime_max_deadline)
+
+int tenant_listen(struct tenant *tenant);
+int listener_thread_register_tenant(struct tenant *tenant);
+void tenant_preprocess(struct http_session *session);
 
 int            tenant_database_add(struct tenant *tenant);
 struct tenant *tenant_database_find_by_name(char *name);
 struct tenant *tenant_database_find_by_socket_descriptor(int socket_descriptor);
 struct tenant *tenant_database_find_by_port(uint16_t port);
 struct tenant *tenant_database_find_by_ptr(void *ptr);
+void           tenant_database_print_reservations();
+void           tenant_database_init_reservations();
+void           tenant_database_replenish_all();
+struct tenant *tenant_database_find_tenant_most_oversupply(struct tenant *tenant_to_exclude, uint64_t time_of_oversupply, bool weak_shed, struct sandbox_metadata **sandbox_meta_to_remove);
 
-typedef void (*tenant_database_foreach_cb_t)(struct tenant *, void *, void *);
-void tenant_database_foreach(tenant_database_foreach_cb_t, void *, void *);
+typedef void (*tenant_database_foreach_cb_t)(struct tenant *, void *);
+void         tenant_database_foreach(tenant_database_foreach_cb_t, void *);
+
+static inline uint64_t
+sandbox_meta_get_priority(void *element)
+{
+	struct sandbox_metadata *sandbox_meta = (struct sandbox_metadata *)element;
+	return sandbox_meta->absolute_deadline;
+}
 
-static inline int
+static inline void
 tenant_policy_specific_init(struct tenant *tenant, struct tenant_config *config)
 {
 	switch (scheduler) {
@@ -31,21 +51,20 @@ tenant_policy_specific_init(struct tenant *tenant, struct tenant_config *config)
 	case SCHEDULER_EDF:
 	case SCHEDULER_SJF:
 		break;
-	case SCHEDULER_MTDS:
+	case SCHEDULER_MTDS: {
 		/* Deferable Server Initialization */
 		tenant->replenishment_period = (uint64_t)config->replenishment_period_us * runtime_processor_speed_MHz;
 		tenant->max_budget           = (uint64_t)config->max_budget_us * runtime_processor_speed_MHz;
 		tenant->remaining_budget     = tenant->max_budget;
 
-		tenant->pwt_sandboxes = (struct perworker_tenant_sandbox_queue *)malloc(
-		  runtime_worker_threads_count * sizeof(struct perworker_tenant_sandbox_queue));
-		if (!tenant->pwt_sandboxes) {
-			fprintf(stderr, "Failed to allocate tenant_sandboxes array: %s\n", strerror(errno));
-			return -1;
-		};
+		config->replenishment_period_us = 0;
+		config->max_budget_us           = 0;
 
-		memset(tenant->pwt_sandboxes, 0,
-		       runtime_worker_threads_count * sizeof(struct perworker_tenant_sandbox_queue));
+		tenant->pwt_sandboxes = (struct perworker_tenant_sandbox_queue *)
+		  calloc(runtime_worker_threads_count, sizeof(struct perworker_tenant_sandbox_queue));
+		if (!tenant->pwt_sandboxes) {
+			panic("Failed to allocate tenant_sandboxes array: %s\n", strerror(errno));
+		}
 
 		for (int i = 0; i < runtime_worker_threads_count; i++) {
 			tenant->pwt_sandboxes[i].sandboxes = priority_queue_initialize(RUNTIME_TENANT_QUEUE_SIZE, false,
@@ -58,20 +77,34 @@ tenant_policy_specific_init(struct tenant *tenant, struct tenant_config *config)
 		}
 
 		/* Initialize the tenant's global request queue */
-		tenant->tgrq_requests                   = malloc(sizeof(struct tenant_global_request_queue));
+		tenant->tgrq_requests = calloc(1, sizeof(struct tenant_global_request_queue));
+		if (!tenant->tgrq_requests) {
+			panic("Failed to allocate tenant global request queue: %s\n", strerror(errno));
+		}
+
 		tenant->tgrq_requests->sandbox_requests = priority_queue_initialize(RUNTIME_TENANT_QUEUE_SIZE, false,
 		                                                                    sandbox_get_priority);
 		tenant->tgrq_requests->tenant           = tenant;
-		tenant->tgrq_requests->mt_class = (tenant->replenishment_period == 0) ? MT_DEFAULT : MT_GUARANTEED;
+		tenant->tgrq_requests->mt_class         = (tenant_is_paid(tenant)) ? MT_GUARANTEED : MT_DEFAULT;
 		tenant->tgrq_requests->tenant_timeout.tenant = tenant;
 		tenant->tgrq_requests->tenant_timeout.pwt    = NULL;
 		break;
-
+	}
 	case SCHEDULER_MTDBF:
+		tenant->reservation_percentile = config->reservation_percentile;
+		config->reservation_percentile = 0;
+
+		ps_list_head_init(&tenant->trs.admitted_jobs_list);
+		ps_list_head_init(&tenant->trs.admitted_BE_jobs_list);
+
+		tenant->local_sandbox_metas =
+		  priority_queue_initialize_new(RUNTIME_RUNQUEUE_SIZE, false, sandbox_meta_get_priority, NULL,
+		                                local_sandbox_meta_update_pq_idx_in_tenant_queue);
+		tenant->global_sandbox_metas =
+		  priority_queue_initialize_new(RUNTIME_RUNQUEUE_SIZE, false, sandbox_meta_get_priority, NULL,
+		                                local_sandbox_meta_update_pq_idx_in_tenant_queue);
 		break;
 	}
-
-	return 0;
 }
 
 static inline struct tenant *
@@ -81,8 +114,9 @@ tenant_alloc(struct tenant_config *config)
 	if (existing_tenant != NULL) panic("Tenant %s is already initialized\n", existing_tenant->name);
 
 	existing_tenant = tenant_database_find_by_port(config->port);
-	if (existing_tenant != NULL)
+	if (existing_tenant != NULL) {
 		panic("Tenant %s is already configured with port %u\n", existing_tenant->name, config->port);
+	}
 
 	struct tenant *tenant = (struct tenant *)calloc(1, sizeof(struct tenant));
 
@@ -91,12 +125,14 @@ tenant_alloc(struct tenant_config *config)
 	tenant->name = config->name;
 	config->name = NULL;
 
+	tenant->max_relative_deadline = (uint64_t)config->max_relative_deadline_us * runtime_processor_speed_MHz;
+
 	tcp_server_init(&tenant->tcp_server, config->port);
 	http_router_init(&tenant->router, config->routes_len);
 	module_database_init(&tenant->module_db);
 	map_init(&tenant->scratch_storage);
 
-	/* Deferrable Server init */
+	/* Scheduling Policy specific tenant init */
 	tenant_policy_specific_init(tenant, config);
 
 	for (int i = 0; i < config->routes_len; i++) {
@@ -176,12 +212,145 @@ get_next_timeout_of_tenant(uint64_t replenishment_period)
 	       + ((now - runtime_boot_timestamp) / replenishment_period + 1) * replenishment_period;
 }
 
+#ifdef TRAFFIC_CONTROL
 
-/**
- * Start the tenant as a server listening at tenant->port
- * @param tenant
- * @returns 0 on success, -1 on error
- */
-int  tenant_listen(struct tenant *tenant);
-int  listener_thread_register_tenant(struct tenant *tenant);
-void tenant_preprocess(struct http_session *session);
+static void
+tenant_print_jobs(struct tenant *tenant)
+{
+	struct tenant_reservation_server *trs  = &tenant->trs;
+	struct job_node                  *head = NULL;
+
+	int i = 0;
+	printf("\nTenant Guaranteed Budget: %lu/%lu\n", trs->budget_guaranteed, trs->max_budget_guaranteed);
+	ps_list_foreach_d(&trs->admitted_jobs_list, head)
+	{
+		printf("GR-Job #%d: Arrival:%lu\t Exec:%lu\n", i++, head->timestamp, head->exec);
+	}
+
+	i = 0;
+	printf("\nTenant Best Effort Budget: %lu/%lu\n", trs->budget_best, UINT64_MAX);
+	ps_list_foreach_d(&trs->admitted_BE_jobs_list, head)
+	{
+		printf("BE-Job #%d: Arrival:%lu\t Exec:%lu\n", i++, head->timestamp, head->exec);
+	}
+}
+
+static void
+tenant_replenish(struct tenant *tenant, uint64_t now)
+{
+	struct tenant_reservation_server *trs  = &tenant->trs;
+	struct job_node                  *head = ps_list_head_first_d(&trs->admitted_jobs_list, struct job_node);
+
+	while (!ps_list_is_head_d(&trs->admitted_jobs_list, head)) {
+		assert(now > head->timestamp);
+		if (now - head->timestamp < REPLENISHMENT_PERIOD) break;
+
+		struct job_node *tmp_next = ps_list_next_d(head);
+
+		trs->budget_guaranteed += head->exec;
+
+		if (head->sandbox_meta) {
+			assert(head->sandbox_meta->trs_job_node);
+			head->sandbox_meta->trs_job_node = NULL;
+		}
+
+		ps_list_rem_d(head);
+		free(head);
+
+		head = tmp_next;
+	}
+
+	assert(trs->budget_guaranteed <= trs->max_budget_guaranteed);
+
+	head = ps_list_head_first_d(&trs->admitted_BE_jobs_list, struct job_node);
+
+	while (!ps_list_is_head_d(&trs->admitted_BE_jobs_list, head)) {
+		assert(now >= head->timestamp);
+		if (now - head->timestamp < REPLENISHMENT_PERIOD) break;
+
+		struct job_node *tmp_next = ps_list_next_d(head);
+
+		trs->budget_best += head->exec;
+
+		ps_list_rem_d(head);
+		free(head);
+
+		head = tmp_next;
+	}
+}
+
+static inline bool
+tenant_can_admit_guaranteed(struct tenant *tenant, uint64_t now, uint64_t adjustment)
+{
+	tenant_replenish(tenant, now);
+	return tenant->trs.budget_guaranteed >= adjustment;
+}
+
+static bool
+tenant_try_add_job_as_guaranteed(struct tenant *tenant, uint64_t arrival_time, uint64_t adjustment, struct sandbox_metadata *sandbox_meta)
+{
+	assert(adjustment > 0);
+
+	struct tenant_reservation_server *trs = &tenant->trs;
+	if (trs->budget_guaranteed < adjustment) return false;
+
+	assert(sandbox_meta->trs_job_node == NULL);
+	assert(sandbox_meta);
+	assert(sandbox_meta->terminated == false);
+	
+	struct job_node *new_node = (struct job_node *)malloc(sizeof(struct job_node));
+	ps_list_init_d(new_node);
+	new_node->exec         = adjustment;
+	new_node->timestamp    = arrival_time;
+	new_node->sandbox_meta = sandbox_meta;
+
+	assert(ps_list_singleton_d(new_node));
+	struct job_node *tail = ps_list_head_last_d(&trs->admitted_jobs_list, struct job_node);
+	ps_list_add_d(tail, new_node);
+	sandbox_meta->trs_job_node = new_node;
+
+	assert(trs->budget_guaranteed >= adjustment);
+	trs->budget_guaranteed -= adjustment;
+	return true;
+}
+
+static void
+tenant_force_add_job_as_best(struct tenant *tenant, uint64_t arrival_time, uint64_t adjustment)
+{
+	assert(adjustment > 0);
+
+	struct tenant_reservation_server *trs = &tenant->trs;
+	struct job_node *new_node = (struct job_node *)malloc(sizeof(struct job_node));
+	ps_list_init_d(new_node);
+	new_node->exec      = adjustment;
+	new_node->timestamp = arrival_time;
+
+	assert(ps_list_singleton_d(new_node));
+	struct job_node *tail = ps_list_head_last_d(&trs->admitted_BE_jobs_list, struct job_node);
+	ps_list_add_d(tail, new_node);
+
+	assert(trs->budget_best > adjustment);
+	trs->budget_best -= adjustment;
+}
+
+static void
+tenant_reduce_guaranteed_job_demand(struct tenant *tenant, uint64_t adjustment, struct sandbox_metadata *sandbox_meta)
+{
+	assert(sandbox_meta);
+	// assert(sandbox_meta->terminated == false); // TODO WHY FIRES???
+	assert(sandbox_meta->global_queue_type == 1);
+	assert(sandbox_meta->trs_job_node);
+
+	struct tenant_reservation_server *trs  = &tenant->trs;
+	struct job_node                  *node = sandbox_meta->trs_job_node;
+	assert(node->sandbox_meta == sandbox_meta);
+
+	assert(node->exec >= adjustment);
+	node->exec -= adjustment;
+	trs->budget_guaranteed += adjustment;
+
+	sandbox_meta->trs_job_node->sandbox_meta = NULL;
+	sandbox_meta->trs_job_node = NULL;
+}
+
+#endif

runtime/include/traffic_control.h

@@ -0,0 +1,16 @@
+#pragma once
+
+#include <stdbool.h>
+#include <stdint.h>
+
+#define TRAFFIC_CONTROL
+// #define LOG_TRAFFIC_CONTROL
+
+typedef struct tenant tenant; // TODO: Why get circular dependency here?
+typedef struct sandbox_metadata sandbox_metadata;
+typedef enum dbf_update_mode dbf_update_mode_t;
+
+void traffic_control_initialize(void);
+void traffic_control_log_decision(const int admissions_case_num, const bool admitted);
+uint64_t  traffic_control_decide(struct sandbox_metadata *sandbox_meta, uint64_t start_time, uint64_t estimated_execution, int *denial_code, int *worker_id_v);
+uint64_t traffic_control_shed_work(struct tenant *tenant_to_exclude, uint64_t time_of_oversupply, int *worker_id_virt_just_shed, bool weak_shed);

runtime/include/wasm_memory.h

@@ -137,9 +137,10 @@ wasm_memory_reinit(struct wasm_memory *wasm_memory, uint64_t initial)
 static INLINE int32_t
 wasm_memory_expand(struct wasm_memory *wasm_memory, uint64_t size_to_expand)
 {
+	assert(wasm_memory);
 	uint64_t target_size = wasm_memory->abi.size + size_to_expand;
 	if (unlikely(target_size > wasm_memory->abi.max)) {
-		fprintf(stderr, "wasm_memory_expand - Out of Memory!. %lu out of %lu\n", wasm_memory->abi.size,
+		fprintf(stderr, "wasm_memory_expand - Out of Memory! target_size=%lu, size=%lu, max=%lu\n", target_size, wasm_memory->abi.size,
 		        wasm_memory->abi.max);
 		return -1;
 	}
@@ -153,6 +154,7 @@ wasm_memory_expand(struct wasm_memory *wasm_memory, uint64_t size_to_expand)
 		int rc = mprotect(wasm_memory->abi.buffer, target_size, PROT_READ | PROT_WRITE);
 		if (rc != 0) {
 			perror("wasm_memory_expand mprotect");
+			assert(0);
 			return -1;
 		}
 
@@ -169,6 +171,9 @@ wasm_memory_get_size(struct wasm_memory *wasm_memory)
 	return wasm_memory->abi.size;
 }
 
+/**
+ * @brief Copy the segments into the linear memory
+ */
 static INLINE void
 wasm_memory_initialize_region(struct wasm_memory *wasm_memory, uint32_t offset, uint32_t region_size, uint8_t region[])
 {
@@ -180,7 +185,7 @@ wasm_memory_initialize_region(struct wasm_memory *wasm_memory, uint32_t offset,
  * instructions. These functions are intended to be used by the runtime to interacts with linear memories. */
 
 /**
- * Translates WASM offsets into runtime VM pointers
+ * Translates WASM offsets into runtime Virtual Memory pointers
  * @param offset an offset into the WebAssembly linear memory
  * @param bounds_check the size of the thing we are pointing to
  * @return void pointer to something in WebAssembly linear memory

runtime/include/wasm_stack.h

@@ -125,6 +125,6 @@ wasm_stack_reinit(struct wasm_stack *wasm_stack)
 	assert(wasm_stack->low == wasm_stack->buffer + /* guard page */ PAGE_SIZE);
 	assert(wasm_stack->high == wasm_stack->low + wasm_stack->capacity);
 
-	explicit_bzero(wasm_stack->low, wasm_stack->capacity);
+	// explicit_bzero(wasm_stack->low, wasm_stack->capacity);
 	ps_list_init_d(wasm_stack);
 }

runtime/include/worker_thread.h

@@ -6,5 +6,6 @@
 
 extern thread_local struct arch_context worker_thread_base_context;
 extern thread_local int                 worker_thread_idx;
+extern thread_local void         *worker_dbf;
 
 void *worker_thread_main(void *return_code);

runtime/src/current_sandbox.c

@@ -68,6 +68,22 @@ current_sandbox_exit()
 	assert(0);
 }
 
+/**
+ * @brief Exit from the executing sandbox during an interrup. 
+ * 
+ * Should be called within the signal handler (preemptive scheduler)
+ * This places the current sandbox on the cleanup queue.
+ */
+void interrupted_sandbox_exit(void)
+{
+	struct sandbox *exiting_sandbox = current_sandbox_get();
+	assert(exiting_sandbox != NULL);
+	assert(exiting_sandbox->state == SANDBOX_INTERRUPTED);
+
+	sandbox_exit_error(exiting_sandbox);
+	local_cleanup_queue_add(exiting_sandbox);
+}
+
 void
 current_sandbox_wasm_trap_handler(int trapno)
 {
@@ -99,7 +115,10 @@ current_sandbox_wasm_trap_handler(int trapno)
 		break;
 	}
 
-	debuglog("%s", error_message);
+	// debuglog("%s - Tenant: %s, Route: %s", error_message, sandbox->tenant->name, sandbox->route->route);
+	debuglog("%s - T: %s, id: %lu, exceeded: %u, rem_exec: %lu, premp: %u, state: %u, abi->size: %lu", error_message, sandbox->tenant->name, sandbox->id,
+		sandbox->exceeded_estimation, sandbox->remaining_exec, sandbox->writeback_preemption_in_progress, sandbox->state, 
+		sandbox->memory->abi.size);
 	current_sandbox_exit();
 	assert(0);
 }
@@ -112,8 +131,8 @@ current_sandbox_init()
 	assert(sandbox != NULL);
 	assert(sandbox->state == SANDBOX_RUNNING_SYS);
 
-	int   rc            = 0;
-	char *error_message = NULL;
+	// int   rc            = 0;
+	// char *error_message = NULL;
 
 	/* Initialize sandbox memory */
 	struct module *current_module = sandbox_get_module(sandbox);
@@ -142,10 +161,10 @@ current_sandbox_init()
 
 	return sandbox;
 
-err:
-	debuglog("%s", error_message);
-	current_sandbox_exit();
-	return NULL;
+// err:
+// 	debuglog("%s", error_message);
+// 	current_sandbox_exit();
+// 	return NULL;
 }
 
 extern noreturn void
@@ -157,8 +176,8 @@ current_sandbox_fini()
 	char *error_message = "";
 	sandbox_syscall(sandbox);
 
-	sandbox->timestamp_of.completion = __getcycles();
-	sandbox->total_time              = sandbox->timestamp_of.completion - sandbox->timestamp_of.allocation;
+	// sandbox->timestamp_of.completion = __getcycles();
+	// sandbox->total_time              = sandbox->timestamp_of.completion - sandbox->timestamp_of.allocation;
 
 	assert(sandbox->state == SANDBOX_RUNNING_SYS);
 
@@ -169,6 +188,7 @@ done:
 	current_sandbox_exit();
 	assert(0);
 err:
+	assert(0);
 	debuglog("%s", error_message);
 	assert(sandbox->state == SANDBOX_RUNNING_SYS);
 
@@ -195,3 +215,26 @@ current_sandbox_start(void)
 
 	if (sandbox->module->type == APP_MODULE) current_sandbox_fini();
 }
+
+int
+sandbox_validate_self_lifetime(struct sandbox *sandbox)
+{
+	if (sandbox->response_code != 0) goto err;
+
+	const uint64_t now = __getcycles();
+	if (sandbox->absolute_deadline >= now + (!sandbox->exceeded_estimation ? sandbox->remaining_exec : 0)) return 0;
+
+	// dbf_try_update_demand(worker_dbf, sandbox->timestamp_of.dispatched,
+	// 		                      sandbox->route->relative_deadline, sandbox->absolute_deadline, sandbox->remaining_exec,
+	// 		                      DBF_DELETE_EXISTING_DEMAND, NULL, NULL);
+
+	assert(sandbox->response_code == 0);
+	sandbox->response_code = 4081;
+
+err:	
+	sandbox_exit_error(sandbox);
+	// sandbox_free(sandbox);
+	// sandbox_free_linear_memory(sandbox);
+	local_cleanup_queue_add(sandbox);
+	return -1;
+}

runtime/src/dbf_array.c.bak

@@ -0,0 +1,335 @@
+#include <string.h>
+#include <assert.h>
+#include "tenant.h"
+#include "runtime.h"
+#include "arch/getcycles.h"
+#include "math.h"
+#include "message.h"
+#include "panic.h"
+#include "dbf.h"
+
+struct tenant;
+
+struct dbf_array {
+	struct tenant *tenant;
+	int      worker_idx;
+	// uint32_t idx_oversupply;
+	uint64_t max_relative_deadline;
+	uint64_t base_supply; /* supply amount for time 1 */
+	uint64_t time_of_oversupply;
+	uint64_t max_absolute_deadline;
+
+	uint32_t capacity;
+	uint64_t demands[];
+};
+
+static inline int 
+dbf_array_get_worker_idx(void * dbf_raw)
+{
+	assert(dbf_raw);
+	struct dbf_array *dbf = (struct dbf_array *)dbf_raw;
+	return dbf->worker_idx;
+}
+
+/*static inline uint64_t
+dbf_array_get_max_relative_dl(void * dbf_raw)
+{
+	assert(dbf_raw);
+	struct dbf_array *dbf = (struct dbf_array *)dbf_raw;
+	return dbf->max_relative_deadline;
+}*/
+
+static inline uint64_t
+dbf_array_get_time_of_oversupply(void * dbf_raw)
+{
+	assert(dbf_raw);
+	struct dbf_array *dbf = (struct dbf_array *)dbf_raw;
+	return dbf->time_of_oversupply;
+}
+
+static void
+dbf_array_print(void *dbf_raw, uint64_t start_time)
+{
+	assert(dbf_raw != NULL);
+	struct dbf_array *dbf = (struct dbf_array *)dbf_raw;
+
+	printf("DBF INFO:\n\
+	\t WorkerIDX: \t%d\n\
+	\t Capacity: \t%u\n\
+	\t Max Rel DL: \t%lu\n\
+	\t Max Abs DL (ms): \t%lu\n\
+	\t Basic Supply: \t%lu\n\n",
+	       dbf->worker_idx, dbf->capacity, dbf->max_relative_deadline, dbf->max_absolute_deadline/runtime_quantum, dbf->base_supply);
+
+	for (int i = 0; i < dbf->capacity; i++) {
+		if (dbf->demands[i] > 0) printf("demands[%d] = %lu\n", i, dbf->demands[i]);
+	}
+}
+
+// static void *
+// dbf_array_grow(void *dbf_raw, uint64_t new_max_relative_deadline)
+// {
+// 	assert(dbf_raw != NULL);
+// 	struct dbf_array *dbf = (struct dbf_array *)dbf_raw;
+
+// 	uint32_t new_capacity = new_max_relative_deadline / runtime_quantum /* * 2 */; // NOT adding 1 for final leftovers
+
+// 	struct dbf_array *new_dbf = realloc(dbf, sizeof(struct dbf_array) + sizeof(uint64_t) * new_capacity);
+// 	if (new_dbf == NULL) panic("Failed to grow dbf\n");
+
+// 	memset(new_dbf->demands, 0, new_capacity * sizeof(uint64_t));
+
+// 	new_dbf->capacity              = new_capacity;
+// 	new_dbf->max_relative_deadline = new_max_relative_deadline;
+
+// 	return new_dbf;
+// }
+
+/*
+static bool
+dbf_array_check_supply_quick(struct dbf_array *dbf_raw, uint64_t start_time, uint64_t abs_deadline, uint64_t adjustment)
+{
+	assert(dbf_raw != NULL);
+	struct dbf_array *dbf = (struct dbf_array *)dbf_raw;
+	// assert(start_time < abs_deadline);
+	if (start_time >= abs_deadline) return true;
+
+	const uint32_t live_deadline_len      = ceil((abs_deadline - start_time) / (double)runtime_quantum);
+	const uint32_t abs_deadline_idx       = (abs_deadline / runtime_quantum) % dbf->capacity;
+	const uint64_t max_supply_at_deadline = live_deadline_len * dbf->base_supply;
+
+	return (dbf->demands[abs_deadline_idx] + adjustment <= max_supply_at_deadline);
+}
+*/
+
+static bool
+dbf_array_try_update_demand(void *dbf_raw, uint64_t start_time, uint64_t route_relative_deadline,
+                      uint64_t abs_deadline, uint64_t adjustment, dbf_update_mode_t dbf_update_mode,
+                      void *new_message_raw, struct sandbox_metadata *sandbox_meta)
+{
+	assert(dbf_raw != NULL);
+	assert(start_time < abs_deadline);
+	struct dbf_array *dbf = (struct dbf_array *)dbf_raw;
+	struct message *new_message = (struct message *) new_message_raw;
+
+	if (abs_deadline > dbf->max_absolute_deadline) dbf->max_absolute_deadline = abs_deadline;
+
+	if (adjustment == 0) goto done;
+
+	// const uint32_t max_relative_deadline_len = dbf->max_relative_deadline / runtime_quantum;
+	const uint32_t live_deadline_len         = round((abs_deadline - start_time) / (double)runtime_quantum);
+	// const uint32_t live_deadline_len         = (abs_deadline - start_time) / runtime_quantum;
+	// const uint32_t live_deadline_len         = abs_deadline/runtime_quantum - start_time/runtime_quantum;
+	const uint32_t abs_deadline_idx          = (abs_deadline / runtime_quantum) % dbf->capacity;
+	// const uint32_t start_time_idx          = (start_time / runtime_quantum) % dbf->capacity;
+	
+	// if (start_time_idx == abs_deadline_idx) goto done;
+
+	// assert(live_deadline_len <= max_relative_deadline_len);
+	assert(live_deadline_len <= dbf->capacity);
+
+	bool demand_is_below_supply = true;
+
+	for (uint32_t i = abs_deadline_idx, iter = 0; i < abs_deadline_idx + dbf->capacity /*iter< (start_time_idx-abs_deadline_idx+dbf->capacity)%dbf->capacity*/; i++, iter++) {
+		uint32_t circular_i = i % dbf->capacity;
+
+		const uint64_t max_supply_at_time_i = (live_deadline_len + iter) * dbf->base_supply;
+		const uint64_t prev_demand          = dbf->demands[circular_i];
+
+		switch (dbf_update_mode) {
+		case DBF_CHECK_AND_ADD_DEMAND:
+			dbf->demands[circular_i] += adjustment;
+
+			if (dbf->demands[circular_i] > max_supply_at_time_i) {
+				/* Undo DBF adding if over supply detected */
+				for (uint32_t j = abs_deadline_idx; j <= i; j++) {
+					dbf->demands[j % dbf->capacity] -= adjustment;
+				}
+				dbf->time_of_oversupply = iter;
+				goto err_demand_over_supply;
+			}
+			break;
+		/*case DBF_CHECK_EXISTING_SANDBOX_EXTRA_DEMAND:
+			if (dbf->demands[circular_i] + adjustment > max_supply_at_time_i) {
+				dbf->time_of_oversupply = iter;
+				goto err_demand_over_supply;
+			}
+			break;*/
+		case DBF_FORCE_ADD_NEW_SANDBOX_DEMAND:
+			/* [Work Conservation Scenario] Only applicable for tenant and global dbf! */
+			assert(dbf->worker_idx < 0);
+
+			dbf->demands[circular_i] += adjustment;
+			assert(prev_demand < dbf->demands[circular_i]);
+
+			if (demand_is_below_supply && dbf->demands[circular_i] > max_supply_at_time_i) {
+				dbf->time_of_oversupply    = iter;
+				demand_is_below_supply = false;
+			}
+			break;
+		case DBF_REDUCE_EXISTING_DEMAND:
+			dbf->demands[circular_i] -= adjustment;
+			if (prev_demand < dbf->demands[circular_i]) {
+				printf("DBF_REDUCE_EXISTING_DEMAND\n");
+				printf("Worker ID: %d\n", dbf->worker_idx);
+				// printf("Tenant Reservation: %u\n", new_message->reserv);
+				printf("Sandbox ID: %lu\n", new_message->sandbox_id);
+				// printf("Sandbox Response Code: %u\n", new_message->sandbox_response_code);
+				printf("Basic supply: %lu\n", dbf->base_supply);
+				printf("Cap=%u\n", dbf->capacity);
+				printf("Abs_dest_idx=%u\n", abs_deadline_idx);
+				printf("live_deadline_len=%u\n", live_deadline_len);
+				printf("i=%u, cir_i = %u, iter = %u\n", i, circular_i, iter);
+				printf("max_supply_at_time_i = %lu\n\n", max_supply_at_time_i);
+				printf("Prev_demand[%u]=%lu\n\n", circular_i, prev_demand);
+				printf("demand[%u]=%lu\n\n", circular_i, dbf->demands[circular_i]);
+				// printf("sandbox_state=%u, if_case=%d\n", new_message->state, new_message->if_case);
+				printf("exceeded_estimation=%d\n", new_message->exceeded_estimation);
+				printf("Adjustment=%lu\n", adjustment);
+				// printf("last_exec_duration=%lu, prev_rem_exec=%ld, rem_exec=%ld\n",
+				//        new_message->last_exec_dur, new_message->prev_rem_exec,
+				//        new_message->remaining_execution);
+
+				dbf_print(dbf, start_time);
+				panic("Interger Underflow -> Tried reducing demand, but it actually went over supply!");
+			}
+			break;
+		}
+	}
+
+done:
+	return demand_is_below_supply;
+err_demand_over_supply:
+	demand_is_below_supply = false;
+	goto done;
+}
+
+static uint64_t
+dbf_array_get_demand_overgone_its_supply_at(void *dbf_raw, uint64_t start_time, uint64_t abs_deadline, uint64_t time_of_oversupply)
+{
+	assert(dbf_raw != NULL);
+	struct dbf_array *dbf = (struct dbf_array *)dbf_raw;
+
+	time_of_oversupply = abs_deadline + time_of_oversupply*runtime_quantum;
+
+	if (time_of_oversupply > dbf->max_absolute_deadline) {
+		printf("abs: %lu, time_of_oversupply: %lu,  dbf_abs: %lu\n", abs_deadline, time_of_oversupply, dbf->max_absolute_deadline);
+		time_of_oversupply = dbf->max_absolute_deadline;
+	}
+
+	assert(start_time < time_of_oversupply);
+
+	// const uint32_t live_deadline_len = (time_of_oversupply - start_time) / runtime_quantum;
+	// const uint32_t live_deadline_len         = time_of_oversupply/runtime_quantum - start_time/runtime_quantum;
+	const uint32_t live_deadline_len         = round((time_of_oversupply - start_time) / (double)runtime_quantum);
+	const uint32_t abs_deadline_idx = (time_of_oversupply / runtime_quantum) % dbf->capacity;
+
+	uint64_t       demand_overgone   = 0;
+	uint32_t circular_i = (time_of_oversupply/runtime_quantum) % dbf->capacity;
+
+	const uint64_t max_supply_at_time_i  = live_deadline_len * dbf->base_supply;
+	const uint64_t curr_demand_at_time_i = dbf->demands[circular_i];
+
+	if (curr_demand_at_time_i > max_supply_at_time_i) {
+		demand_overgone = curr_demand_at_time_i - max_supply_at_time_i;
+	}
+
+	return demand_overgone;
+}
+
+/*
+static uint64_t
+dbf_array_get_demand_overgone_its_supply_at(void *dbf_raw, uint64_t start_time, uint64_t abs_deadline, uint64_t time_of_oversupply)
+{
+	assert(dbf_raw != NULL);
+	assert(start_time < abs_deadline);
+	struct dbf_array *dbf = (struct dbf_array *)dbf_raw;
+
+	const uint32_t live_deadline_len = ceil((abs_deadline - start_time) / (double)runtime_quantum);
+	const uint32_t abs_deadline_idx = (abs_deadline / runtime_quantum) % dbf->capacity;
+
+	uint64_t       demand_overgone   = 0;
+	uint32_t circular_i = (abs_deadline_idx + time_of_oversupply) % dbf->capacity;
+
+	const uint64_t max_supply_at_time_i  = (live_deadline_len + time_of_oversupply) * dbf->base_supply;
+	const uint64_t curr_demand_at_time_i = dbf->demands[circular_i];
+
+	if (curr_demand_at_time_i > max_supply_at_time_i) {
+		demand_overgone = curr_demand_at_time_i - max_supply_at_time_i;
+	}
+
+	return demand_overgone;
+}
+
+static uint64_t
+dbf_array_get_demand_overgone_its_supply_at__BAK(void *dbf_raw, uint64_t start_time, uint64_t abs_deadline)
+{
+	assert(dbf_raw != NULL);
+	assert(start_time < abs_deadline);
+	struct dbf_array *dbf = (struct dbf_array *)dbf_raw;
+
+	const uint32_t live_deadline_len    = ceil((abs_deadline - start_time) / (double)runtime_quantum);
+	const uint32_t absolute_arrival_idx = start_time / runtime_quantum % dbf->capacity;
+
+	uint64_t demand_overgone = 0;
+
+	const uint32_t abs_deadline_idx = (abs_deadline / runtime_quantum) % dbf->capacity;
+
+	// assert(live_deadline_len<=route_relative_deadline_len);
+
+	for (uint32_t i = abs_deadline_idx, iter = 0; i < abs_deadline_idx + live_deadline_len; i++, iter++) {
+		uint32_t circular_i = i % dbf->capacity;
+
+		const uint64_t max_supply_at_time_i  = (live_deadline_len + iter) * dbf->base_supply;
+		const uint64_t curr_demand_at_time_i = dbf->demands[circular_i];
+
+		if (curr_demand_at_time_i > max_supply_at_time_i) {
+			if (curr_demand_at_time_i - max_supply_at_time_i > demand_overgone) {
+				demand_overgone = curr_demand_at_time_i - max_supply_at_time_i;
+			}
+		}
+	}
+
+	return demand_overgone;
+}
+*/
+static void
+dbf_array_free(void *dbf)
+{
+	assert(dbf != NULL);
+
+	free(dbf);
+}
+
+
+void *
+dbf_array_initialize(uint32_t num_of_workers, uint8_t reservation_percentile, int worker_idx, struct tenant *tenant)
+{
+	struct dbf_config config = {
+		.get_worker_idx_fn = dbf_array_get_worker_idx,
+		// .get_max_relative_dl_fn = dbf_array_get_max_relative_dl,
+		// .get_idx_oversuplly_fn = dbf_array_get_idx_oversuplly,
+		.get_time_of_oversupply_fn = dbf_array_get_time_of_oversupply,
+		.print_fn               = dbf_array_print,
+		// .print_supply_fn            = dbf_array_print_suply,
+		// .grow_fn = 								dbf_array_grow,
+		.try_update_demand_fn                 = dbf_array_try_update_demand,
+		.get_demand_overgone_its_supply_at_fn = dbf_array_get_demand_overgone_its_supply_at,
+		.free_fn = dbf_array_free
+	};
+
+	dbf_plug_functions(&config);
+
+	assert(runtime_max_deadline > 0);
+
+	uint32_t capacity = runtime_max_deadline / runtime_quantum /* * 2 */; // NOT adding 1 for final leftovers
+	struct dbf_array *dbf = (struct dbf_array *)calloc(1, sizeof(struct dbf_array) + sizeof(uint64_t) * capacity);
+
+	dbf->capacity              = capacity;
+	dbf->max_relative_deadline = runtime_max_deadline;
+	dbf->worker_idx            = worker_idx;
+	// uint32_t cpu_factor        = (num_of_workers == 1) ? 1 : num_of_workers * RUNTIME_MAX_CPU_UTIL_PERCENTILE / 100;
+	dbf->base_supply           = runtime_quantum * num_of_workers * reservation_percentile * RUNTIME_MAX_CPU_UTIL_PERCENTILE / 10000;
+
+	return dbf;
+}

runtime/src/dbf_generic.c

@@ -0,0 +1,80 @@
+#include <string.h>
+#include <assert.h>
+#include "dbf.h"
+
+static struct dbf_config dbf_conf;
+// void *global_dbf_temp;
+
+int
+dbf_get_worker_idx(void *dbf)
+{
+	assert(dbf_conf.get_worker_idx_fn != NULL);
+	return dbf_conf.get_worker_idx_fn(dbf);
+}
+
+// uint64_t
+// dbf_get_max_relative_dl(void *dbf)
+// {
+// 	assert(dbf_conf.get_max_relative_dl_fn != NULL);
+// 	return dbf_conf.get_max_relative_dl_fn(dbf);
+// }
+
+uint64_t
+dbf_get_time_of_oversupply(void *dbf)
+{
+	assert(dbf_conf.get_time_of_oversupply_fn != NULL);
+	return dbf_conf.get_time_of_oversupply_fn(dbf);
+}
+
+void
+dbf_print(void *dbf, uint64_t start_time)
+{
+	assert(dbf_conf.print_fn != NULL);
+	return dbf_conf.print_fn(dbf, start_time);
+}
+
+// void *
+// dbf_grow(void *dbf, uint64_t new_max_relative_deadline)
+// {
+// 	assert(dbf_conf.grow_fn != NULL);
+// 	return dbf_conf.grow_fn(dbf, new_max_relative_deadline);
+// }
+
+bool
+dbf_try_update_demand(void *dbf, uint64_t start_time, uint64_t route_relative_deadline, uint64_t abs_deadline,
+                      uint64_t adjustment, dbf_update_mode_t dbf_update_mode, void *new_message, struct sandbox_metadata *sandbox_meta)
+{
+	assert(dbf_conf.try_update_demand_fn != NULL);
+	return dbf_conf.try_update_demand_fn(dbf, start_time, route_relative_deadline, abs_deadline, adjustment,
+	                                     dbf_update_mode, new_message, sandbox_meta);
+}
+
+uint64_t
+dbf_get_demand_overgone_its_supply_at(void *dbf, uint64_t start_time, uint64_t abs_deadline, uint64_t time_of_oversupply)
+{
+	assert(dbf_conf.get_demand_overgone_its_supply_at_fn != NULL);
+	return dbf_conf.get_demand_overgone_its_supply_at_fn(dbf, start_time, abs_deadline, time_of_oversupply);
+}
+
+void
+dbf_free(void *dbf)
+{
+	assert(dbf_conf.free_fn != NULL);
+	return dbf_conf.free_fn(dbf);
+}
+
+void
+dbf_plug_functions(struct dbf_config *config)
+{
+	memcpy(&dbf_conf, config, sizeof(struct dbf_config));
+}
+
+void 
+*dbf_initialize(uint32_t num_of_workers, uint8_t reservation_percentile, int worker_idx, struct tenant *tenant)
+{
+#ifdef DBF_USE_LINKEDLIST
+	return dbf_list_initialize(num_of_workers, reservation_percentile, worker_idx, tenant);
+#else
+	return dbf_array_initialize(num_of_workers, reservation_percentile, worker_idx, tenant);
+#endif
+}

runtime/src/dbf_list.c

@@ -0,0 +1,199 @@
+#include <assert.h>
+#include "dbf.h"
+#include "sandbox_types.h"
+
+struct dbf_list {
+	struct tenant *tenant;
+	int      worker_idx;
+	uint64_t max_relative_deadline;
+	double base_supply; /* supply amount for time 1 */
+	uint64_t time_of_oversupply;
+	uint64_t demand_total;
+
+	struct ps_list_head demands_list;
+};
+
+static inline int 
+dbf_list_get_worker_idx(void * dbf_raw)
+{
+	assert(dbf_raw);
+	struct dbf_list *dbf = (struct dbf_list *)dbf_raw;
+	return dbf->worker_idx;
+}
+
+/*static inline uint64_t
+dbf_list_get_max_relative_dl(void * dbf_raw)
+{
+	assert(dbf_raw);
+	struct dbf_list *dbf = (struct dbf_list *)dbf_raw;
+	return dbf->max_relative_deadline;
+}*/
+
+static inline uint64_t
+dbf_list_get_time_of_oversupply(void * dbf_raw)
+{
+	assert(dbf_raw);
+	struct dbf_list *dbf = (struct dbf_list *)dbf_raw;
+	return dbf->time_of_oversupply;
+}
+
+static void
+dbf_list_print(void *dbf_raw, uint64_t start_time)
+{
+	assert(dbf_raw != NULL);
+	struct dbf_list *dbf = (struct dbf_list *)dbf_raw;
+
+	printf("DBF INFO LL:\n\
+	\t WorkerIDX: \t%d\n\
+	\t Basic Supply: \t%lf\n\n", dbf->worker_idx, dbf->base_supply);
+
+	struct demand_node *node = NULL;
+	uint64_t demand_sum = 0;
+
+	ps_list_foreach_d(&dbf->demands_list, node) 
+	{
+		const uint32_t live_deadline_len = node->abs_deadline - start_time;
+		const uint64_t max_supply_at_time_i = live_deadline_len * dbf->base_supply;
+		demand_sum += node->demand;
+		uint64_t over = 0;
+		if (demand_sum >= max_supply_at_time_i) over = demand_sum - max_supply_at_time_i;
+		printf("demand_at[%lu] = %lu, t=%s, demand_sum=%lu/supply=%lu, demand_over=%lu\n", node->abs_deadline, node->demand, node->tenant->name, demand_sum, max_supply_at_time_i, over);
+	}
+}
+
+bool
+dbf_list_try_add_new_demand(void *dbf_raw, uint64_t start_time, uint64_t abs_deadline, uint64_t adjustment, struct sandbox_metadata *sm)
+{
+	assert(dbf_raw != NULL);
+	assert(start_time < abs_deadline);
+	assert(sm);
+	assert(sm->demand_node == NULL);
+	assert(adjustment > 0);
+	// if (adjustment == 0) return false;
+
+	struct dbf_list *dbf = (struct dbf_list *)dbf_raw;
+	struct demand_node *node = NULL;
+	uint64_t past_deadline_demand = 0;
+	uint64_t demand_sum = 0;
+
+	ps_list_foreach_d(&dbf->demands_list, node) 
+	{
+		if (node->abs_deadline <= start_time) past_deadline_demand = demand_sum;
+		else if (node->abs_deadline >= abs_deadline) break;
+		demand_sum += node->demand;
+	}
+
+	struct demand_node *node_spot = node;
+	assert(abs_deadline != node->abs_deadline);
+	assert(abs_deadline == sm->absolute_deadline);
+
+	demand_sum += adjustment;
+	const uint64_t live_deadline_len = abs_deadline - start_time;
+	const uint64_t max_supply_at_time_i = live_deadline_len * dbf->base_supply; // + past_deadline_demand;
+	if (demand_sum > max_supply_at_time_i) {
+		dbf->time_of_oversupply = abs_deadline;
+		goto err;
+	}
+
+	while(!ps_list_is_head_d(&dbf->demands_list, node)) {
+		struct demand_node *tmp_next = ps_list_next_d(node);
+		const uint64_t live_deadline_len = node->abs_deadline - start_time;
+		const uint64_t max_supply_at_time_i = live_deadline_len * dbf->base_supply; // + past_deadline_demand;
+		demand_sum += node->demand;
+		if (demand_sum > max_supply_at_time_i) {
+			dbf->time_of_oversupply = node->abs_deadline;
+			goto err;
+		}
+		node = tmp_next;
+	}
+
+	struct demand_node *new_node = (struct demand_node*) malloc(sizeof(struct demand_node));
+	ps_list_init_d(new_node);
+	new_node->abs_deadline = abs_deadline;
+	new_node->demand = adjustment;
+	new_node->tenant = sm->tenant;
+	// new_node->sandbox_meta = sm;
+	sm->demand_node = new_node;
+	assert(ps_list_singleton_d(new_node));
+	ps_list_append_d(node_spot, new_node);
+	dbf->demand_total = demand_sum + adjustment;
+	return true;
+err:
+	return false;
+}
+
+void
+dbf_list_force_add_extra_slack(void *dbf_raw, struct sandbox_metadata *sm, uint64_t adjustment)
+{
+	assert(dbf_raw != NULL);
+	assert(sm);
+	assert(sm->demand_node);
+	assert(adjustment > 0);
+
+	struct demand_node *node = sm->demand_node;
+	assert(node->abs_deadline == sm->absolute_deadline);
+	assert(node->demand >= adjustment);
+	node->demand += adjustment;
+
+	struct dbf_list *dbf = (struct dbf_list *)dbf_raw;
+	dbf->demand_total += adjustment;
+}
+
+void
+dbf_list_reduce_demand(struct sandbox_metadata *sm, uint64_t adjustment, bool delete_node)
+{
+	assert(sm);
+	assert(sm->demand_node);
+	assert(delete_node || adjustment > 0);
+
+	struct demand_node *node = sm->demand_node;
+	assert(node->abs_deadline == sm->absolute_deadline);
+	assert(node->demand >= adjustment);
+	node->demand -= adjustment;
+
+	// assert(dbf->demand_total >= adjustment);
+	// dbf->demand_total -= adjustment;
+
+	if (delete_node) {
+		assert(node->demand == 0);
+		/* Clean up empty and repetitive nodes */
+		ps_list_rem_d(node);
+		free(node);
+		node = NULL;
+	}
+}
+
+static void
+dbf_list_free(void *dbf)
+{
+	assert(dbf != NULL);
+
+	free(dbf);
+}
+
+void *
+dbf_list_initialize(uint32_t num_of_workers, uint8_t reservation_percentile, int worker_idx, struct tenant *tenant)
+{
+	struct dbf_config config = {
+		// .try_update_demand_fn = dbf_list_try_add_new_demand,
+		.get_worker_idx_fn = dbf_list_get_worker_idx,
+		.get_time_of_oversupply_fn = dbf_list_get_time_of_oversupply,
+		.print_fn               = dbf_list_print,
+		.free_fn = dbf_list_free
+	};
+
+	dbf_plug_functions(&config);
+
+	assert(runtime_max_deadline > 0);
+	
+	struct dbf_list *dbf = (struct dbf_list *)calloc(1, sizeof(struct dbf_list));
+	ps_list_head_init(&dbf->demands_list);
+
+	dbf->max_relative_deadline = runtime_max_deadline;
+	dbf->worker_idx            = worker_idx;
+	// uint32_t cpu_factor        = (num_of_workers == 1) ? 1 : num_of_workers * RUNTIME_MAX_CPU_UTIL_PERCENTILE / 100;
+	dbf->base_supply           = /*runtime_quantum * */1.0*num_of_workers * reservation_percentile * RUNTIME_MAX_CPU_UTIL_PERCENTILE / 10000;
+	dbf->tenant = tenant;
+
+	return dbf;
+}

runtime/src/global_request_scheduler.c

@@ -3,6 +3,8 @@
 #include "global_request_scheduler.h"
 #include "panic.h"
 
+static struct sandbox_metadata global_highest_priority_metadata;
+
 /* Default uninitialized implementations of the polymorphic interface */
 noreturn static struct sandbox *
 uninitialized_add(struct sandbox *arg)
@@ -87,3 +89,45 @@ global_request_scheduler_peek()
 {
 	return global_request_scheduler.peek_fn();
 }
+
+/**
+ * Peeks at the metadata of the highest priority sandbox
+ * @returns metadata of the highest priority sandbox
+ */
+struct sandbox_metadata
+global_request_scheduler_peek_metadata()
+{
+	return global_highest_priority_metadata;
+}
+
+/**
+ * Updates the metadata of the highest priority sandbox
+ * @param element the highest priority sandbox
+ */
+void
+global_request_scheduler_update_highest_priority(const void *element)
+{
+	if (element == NULL) {
+		global_highest_priority_metadata.absolute_deadline    = UINT64_MAX;
+		global_highest_priority_metadata.tenant               = NULL;
+		global_highest_priority_metadata.route                = NULL;
+		global_highest_priority_metadata.allocation_timestamp = 0;
+		global_highest_priority_metadata.remaining_exec  = 0;
+		global_highest_priority_metadata.id                   = 0;
+		global_highest_priority_metadata.global_queue_type    = 0;
+		global_highest_priority_metadata.exceeded_estimation  = false;
+		global_highest_priority_metadata.state                = SANDBOX_UNINITIALIZED;
+		return;
+	}
+
+	const struct sandbox *sandbox                         = element;
+	global_highest_priority_metadata.absolute_deadline    = sandbox->absolute_deadline;
+	global_highest_priority_metadata.tenant               = sandbox->tenant;
+	global_highest_priority_metadata.route                = sandbox->route;
+	global_highest_priority_metadata.allocation_timestamp = sandbox->timestamp_of.allocation;
+	global_highest_priority_metadata.remaining_exec  = sandbox->remaining_exec;
+	// global_highest_priority_metadata.remaining_exec  = sandbox->remaining_execution_original;
+	global_highest_priority_metadata.id                   = sandbox->id;
+	global_highest_priority_metadata.exceeded_estimation  = sandbox->exceeded_estimation;
+	global_highest_priority_metadata.state                = sandbox->state;
+}

runtime/src/global_request_scheduler_minheap.c

@@ -61,23 +61,13 @@ global_request_scheduler_minheap_peek(void)
 	return priority_queue_peek(global_request_scheduler_minheap);
 }
 
-uint64_t
-sandbox_get_priority_fn(void *element)
-{
-	struct sandbox *sandbox = (struct sandbox *)element;
-	if (scheduler == SCHEDULER_SJF) return sandbox->remaining_exec;
-	assert(scheduler == SCHEDULER_EDF);
-	return sandbox->absolute_deadline;
-};
-
-
 /**
  * Initializes the variant and registers against the polymorphic interface
  */
 void
 global_request_scheduler_minheap_initialize()
 {
-	global_request_scheduler_minheap = priority_queue_initialize(4096, true, sandbox_get_priority_fn);
+	global_request_scheduler_minheap = priority_queue_initialize(4096, true, sandbox_get_priority);
 
 	struct global_request_scheduler_config config = {.add_fn    = global_request_scheduler_minheap_add,
 	                                                 .remove_fn = global_request_scheduler_minheap_remove,

runtime/src/global_request_scheduler_mtdbf.c

@@ -0,0 +1,204 @@
+#include <assert.h>
+#include <errno.h>
+
+#include "global_request_scheduler.h"
+#include "listener_thread.h"
+#include "panic.h"
+#include "priority_queue.h"
+#include "runtime.h"
+#include "tenant_functions.h"
+#include "sandbox_set_as_error.h"
+#include "dbf.h"
+#include "local_cleanup_queue.h"
+
+struct priority_queue                *global_request_scheduler_mtdbf;
+
+lock_t global_lock;
+// int max_global_runqueue_len = 0; //////////
+
+/**
+ * Pushes a sandbox request to the global runqueue
+ * @param sandbox
+ * @returns pointer to request if added. NULL otherwise
+ */
+static struct sandbox *
+global_request_scheduler_mtdbf_add(struct sandbox *sandbox)
+{
+	assert(sandbox);
+	assert(global_request_scheduler_mtdbf);
+	assert(listener_thread_is_running());
+
+	lock_node_t node = {};
+	lock_lock(&global_lock, &node);
+
+	int rc = priority_queue_enqueue_nolock(global_request_scheduler_mtdbf, sandbox);
+	if (rc != 0) {
+		assert(sandbox->response_code == 0);
+		sandbox->response_code = 4293;
+		sandbox = NULL; // TODO: FIX ME
+		goto done;
+	} 
+	
+	sandbox->owned_worker_idx = -1;
+
+	// if(priority_queue_length_nolock(global_request_scheduler_mtdbf) > max_global_runqueue_len) {
+	// 	max_global_runqueue_len = priority_queue_length_nolock(global_request_scheduler_mtdbf);
+	// 	printf("Global MAX Queue Length: %u\n", max_global_runqueue_len);
+	// }
+	// printf("GlobalLen: %d, Tenant: %s, Tenant-G: %d, Tenant-L: %d\n\n", priority_queue_length_nolock(global_request_scheduler_mtdbf), sandbox->tenant->name, 
+	// 	priority_queue_length_nolock(sandbox->tenant->global_sandbox_metas), priority_queue_length_nolock(sandbox->tenant->local_sandbox_metas));
+
+done: 
+	lock_unlock(&global_lock, &node);
+	return sandbox;
+}
+
+/**
+ * @param pointer to the pointer that we want to set to the address of the removed sandbox request
+ * @returns 0 if successful, -ENOENT if empty
+ */
+int
+global_request_scheduler_mtdbf_remove(struct sandbox **removed_sandbox)
+{
+	/* This function won't be used with the MTDS scheduler. Keeping merely for the polymorhism. */
+	return -1;
+}
+
+/**
+ * @param removed_sandbox pointer to set to removed sandbox request
+ * @param target_deadline the deadline that the request must be earlier than to dequeue
+ * @returns 0 if successful, -ENOENT if empty or if request isn't earlier than target_deadline
+ */
+int
+global_request_scheduler_mtdbf_remove_if_earlier(struct sandbox **removed_sandbox, uint64_t target_deadline)
+{
+	int rc = -ENOENT;
+
+	const uint64_t now = __getcycles();
+	struct sandbox *local = local_runqueue_get_next();
+
+	uint64_t local_rem = local == NULL ? 0 : local->remaining_exec;
+
+	lock_node_t node = {};
+	lock_lock(&global_lock, &node);
+
+	struct sandbox_metadata global_metadata = global_request_scheduler_peek_metadata();
+	uint64_t global_deadline = global_metadata.absolute_deadline;
+
+	if(USING_EARLIEST_START_FIRST) {
+		if (global_deadline - global_metadata.remaining_exec >= target_deadline - local_rem) goto err_enoent;
+	} else {
+		if (global_deadline >= target_deadline) goto err_enoent;
+	}
+	// if (global_deadline == UINT64_MAX) goto err_enoent;
+
+	/* Spot the sandbox to remove */
+	struct sandbox *top_sandbox = NULL;
+	rc = priority_queue_top_nolock(global_request_scheduler_mtdbf, (void **)&top_sandbox);
+	assert(top_sandbox);
+	assert(top_sandbox->absolute_deadline == global_deadline);
+	assert(top_sandbox->remaining_exec == global_metadata.remaining_exec);
+	assert(top_sandbox->state == SANDBOX_INITIALIZED || top_sandbox->state == SANDBOX_PREEMPTED);
+	assert(top_sandbox->response_code == 0);
+
+	if (top_sandbox->sandbox_meta->terminated) {
+		assert(top_sandbox->sandbox_meta->error_code > 0);
+		top_sandbox->response_code = top_sandbox->sandbox_meta->error_code;
+	} else if (global_deadline < now + (!top_sandbox->exceeded_estimation ? top_sandbox->remaining_exec : 0)) {
+		top_sandbox->response_code = top_sandbox->state == SANDBOX_INITIALIZED ? 4080 : 4082;
+	} else if (USING_LOCAL_RUNQUEUE) {
+		struct tenant *tenant       = top_sandbox->tenant;
+		struct route  *route        = top_sandbox->route;
+
+		// assert(dbf_get_worker_idx(worker_dbf) == worker_thread_idx);
+		// if (!dbf_try_update_demand(worker_dbf, now, route->relative_deadline,
+	    //                        global_deadline, top_sandbox->remaining_exec, DBF_CHECK_AND_ADD_DEMAND, NULL, NULL)) {
+		// 	goto err_enoent;
+		// }
+	}
+	else if(local) {
+		assert(USING_WRITEBACK_FOR_PREEMPTION);
+		assert(local->state == SANDBOX_INTERRUPTED);
+		assert(local->writeback_preemption_in_progress == false);
+		assert(local->owned_worker_idx >= 0);
+		assert(local->pq_idx_in_runqueue >= 1);
+		local->writeback_preemption_in_progress = true;
+		local_runqueue_delete(local);
+		// local->response_code = 5000;
+		// interrupted_sandbox_exit();
+	} 
+
+	top_sandbox->timestamp_of.dispatched = now; // remove the same op from scheduler validate and set_as_runable 
+	top_sandbox->owned_worker_idx = -2;
+	// printf("Worker %i accepted a sandbox #%lu!\n", worker_thread_idx, top_sandbox->id);
+
+	rc = priority_queue_dequeue_nolock(global_request_scheduler_mtdbf, (void **)removed_sandbox);
+	assert(rc == 0);
+	assert(*removed_sandbox == top_sandbox);
+
+	assert(top_sandbox->state == SANDBOX_INITIALIZED || top_sandbox->state == SANDBOX_PREEMPTED);
+
+	lock_unlock(&global_lock, &node);
+
+done:
+	return rc;
+err_enoent:
+	lock_unlock(&global_lock, &node);
+	rc = -ENOENT;
+	goto done;
+}
+
+/**
+ * @param removed_sandbox pointer to set to removed sandbox request
+ * @param target_deadline the deadline that the request must be earlier than to dequeue
+ * @param mt_class the multi-tenancy class of the global request to compare the target deadline against
+ * @returns 0 if successful, -ENOENT if empty or if request isn't earlier than target_deadline
+ */
+int
+global_request_scheduler_mtdbf_remove_with_mt_class(struct sandbox **removed_sandbox, uint64_t target_deadline,
+                                                    enum MULTI_TENANCY_CLASS target_mt_class)
+{
+	/* This function won't be used with the MTDBF scheduler. Keeping merely for the polymorhism. */
+	return -1;
+}
+
+/**
+ * Peek at the priority of the highest priority task without having to take the lock
+ * Because this is a min-heap PQ, the highest priority is the lowest 64-bit integer
+ * This is used to store an absolute deadline
+ * @returns value of highest priority value in queue or ULONG_MAX if empty
+ */
+static uint64_t
+global_request_scheduler_mtdbf_peek(void)
+{
+	return priority_queue_peek(global_request_scheduler_mtdbf);
+}
+
+
+/**
+ * Initializes the variant and registers against the polymorphic interface
+ */
+void
+global_request_scheduler_mtdbf_initialize()
+{
+	global_request_scheduler_mtdbf = priority_queue_initialize_new(RUNTIME_RUNQUEUE_SIZE, false, USING_EARLIEST_START_FIRST ? sandbox_get_priority_global : sandbox_get_priority,
+	                                                               global_request_scheduler_update_highest_priority,
+	                                                               sandbox_update_pq_idx_in_runqueue);
+
+	lock_init(&global_lock);
+
+	struct global_request_scheduler_config config = {
+		.add_fn               = global_request_scheduler_mtdbf_add,
+		.remove_fn            = global_request_scheduler_mtdbf_remove,
+		.remove_if_earlier_fn = global_request_scheduler_mtdbf_remove_if_earlier,
+		.peek_fn              = global_request_scheduler_mtdbf_peek
+	};
+
+	global_request_scheduler_initialize(&config);
+}
+
+void
+global_request_scheduler_mtdbf_free()
+{
+	priority_queue_free(global_request_scheduler_mtdbf);
+}

runtime/src/libc/wasi_impl_serverless.c

@@ -1040,6 +1040,7 @@ wasi_snapshot_preview1_backing_poll_oneoff(wasi_context_t *context, const __wasi
 noreturn void
 wasi_snapshot_preview1_backing_proc_exit(wasi_context_t *context, __wasi_exitcode_t exitcode)
 {
+	// panic("This path should not be reachable\n");
 	current_sandbox_fini();
 	assert(0);
 }

runtime/src/listener_thread.c

@@ -1,6 +1,8 @@
 #include <stdint.h>
 #include <unistd.h>
 
+#include "admissions_control.h"
+#include "traffic_control.h"
 #include "arch/getcycles.h"
 #include "execution_regression.h"
 #include "global_request_scheduler.h"
@@ -15,6 +17,13 @@
 #include "tenant.h"
 #include "tenant_functions.h"
 
+#include "sandbox_perf_log.h"
+#include "http_session_perf_log.h"
+#include "ck_ring.h"
+#include "priority_queue.h"
+#include "global_request_scheduler_mtdbf.h"
+#include "sandbox_set_as_error.h"
+
 static void listener_thread_unregister_http_session(struct http_session *http);
 static void panic_on_epoll_error(struct epoll_event *evt);
 
@@ -35,6 +44,11 @@ int listener_thread_epoll_file_descriptor;
 
 pthread_t listener_thread_id;
 
+struct comm_with_worker *comm_from_workers, *comm_to_workers;
+extern lock_t global_lock;
+
+static struct sandbox_metadata *global_sandbox_meta = NULL;
+
 /**
  * Initializes the listener thread, pinned to core 0, and starts to listen for requests
  */
@@ -42,6 +56,12 @@ void
 listener_thread_initialize(void)
 {
 	printf("Starting listener thread\n");
+
+	comm_from_workers       = calloc(runtime_worker_threads_count, sizeof(struct comm_with_worker));
+	comm_to_workers         = calloc(runtime_worker_threads_count, sizeof(struct comm_with_worker));
+	comm_with_workers_init(comm_from_workers);
+	comm_with_workers_init(comm_to_workers);
+
 	cpu_set_t cs;
 
 	CPU_ZERO(&cs);
@@ -55,14 +75,14 @@ listener_thread_initialize(void)
 	assert(ret == 0);
 	ret = pthread_setaffinity_np(listener_thread_id, sizeof(cpu_set_t), &cs);
 	assert(ret == 0);
-	ret = pthread_setaffinity_np(pthread_self(), sizeof(cpu_set_t), &cs);
+	if (geteuid() != 0) ret = pthread_setaffinity_np(pthread_self(), sizeof(cpu_set_t), &cs);
 	assert(ret == 0);
 
 	printf("\tListener core thread: %lx\n", listener_thread_id);
 }
 
 /**
- * @brief Registers a serverless tenant on the listener thread's epoll descriptor
+ * @brief Registers a serverless http session on the listener thread's epoll descriptor
  **/
 void
 listener_thread_register_http_session(struct http_session *http)
@@ -99,7 +119,7 @@ listener_thread_register_http_session(struct http_session *http)
 }
 
 /**
- * @brief Registers a serverless tenant on the listener thread's epoll descriptor
+ * @brief Registers a serverless http session on the listener thread's epoll descriptor
  **/
 static void
 listener_thread_unregister_http_session(struct http_session *http)
@@ -153,6 +173,402 @@ listener_thread_register_metrics_server()
 	return rc;
 }
 
+static void
+check_messages_from_workers()
+{
+#ifdef TRAFFIC_CONTROL
+	assert(comm_from_workers);
+	assert(comm_to_workers);
+
+	for (int worker_idx = 0; worker_idx < runtime_worker_threads_count; worker_idx++) {
+		struct message           new_message = { 0 };
+		struct comm_with_worker *cfw         = &comm_from_workers[worker_idx];
+		struct comm_with_worker *ctw         = &comm_to_workers[worker_idx];
+		assert(cfw);
+		assert(ctw);
+		assert(cfw->worker_idx == worker_idx);
+		assert(ctw->worker_idx == worker_idx);
+
+		const uint64_t now = __getcycles();
+
+		while (ck_ring_dequeue_spsc_message(&cfw->worker_ring, cfw->worker_ring_buffer, &new_message)) {
+			assert(new_message.sender_worker_idx == worker_idx);
+			assert(new_message.sandbox_meta);
+			
+			struct sandbox_metadata *sandbox_meta = new_message.sandbox_meta;
+			assert(new_message.sandbox_id == sandbox_meta->id);
+
+			struct tenant *tenant = sandbox_meta->tenant;
+			struct route *route = sandbox_meta->route;
+			const uint64_t absolute_deadline = sandbox_meta->absolute_deadline;		
+			const uint64_t allocation_timestamp = sandbox_meta->allocation_timestamp;
+			
+			assert(tenant);
+			assert(route);
+			assert(absolute_deadline > 0);
+			assert(allocation_timestamp > 0);
+
+			sandbox_meta->exceeded_estimation = new_message.exceeded_estimation;
+			sandbox_meta->state = new_message.state;
+			sandbox_meta->owned_worker_idx = new_message.sender_worker_idx;
+
+			switch (new_message.message_type)
+			{
+			case MESSAGE_CFW_PULLED_NEW_SANDBOX: {
+				assert(sandbox_meta->state == SANDBOX_RUNNABLE);
+
+				if (sandbox_meta->terminated) continue;
+				
+				if (sandbox_meta->pq_idx_in_tenant_queue) {
+					assert(sandbox_meta->global_queue_type == 2);
+					assert(sandbox_meta->tenant_queue == tenant->global_sandbox_metas);
+					priority_queue_delete_by_idx_nolock(tenant->global_sandbox_metas,
+											sandbox_meta,sandbox_meta->pq_idx_in_tenant_queue);
+					if(unlikely(priority_queue_enqueue_nolock(tenant->local_sandbox_metas, sandbox_meta))){
+						panic("Failed to add sandbox_meta to tenant metadata queue");
+					}
+					sandbox_meta->tenant_queue = tenant->local_sandbox_metas;
+				}
+				break;
+			}
+			case MESSAGE_CFW_DELETE_SANDBOX: {
+				assert(sandbox_meta->state == SANDBOX_RETURNED || sandbox_meta->state == SANDBOX_ERROR);
+				// assert(new_message.adjustment > 0);
+				assert(sandbox_meta->terminated || sandbox_meta->remaining_exec > 0);
+				if (!sandbox_meta->terminated){					
+					assert(sandbox_meta->worker_id_virt >= 0);
+					// assert(sandbox_meta->remaining_exec == new_message.adjustment);
+					void *global_dbf = global_virt_worker_dbfs[sandbox_meta->worker_id_virt];
+
+					dbf_list_reduce_demand(sandbox_meta, sandbox_meta->remaining_exec + sandbox_meta->extra_slack, true);
+					sandbox_meta->demand_node = NULL;
+					// sandbox_meta->extra_slack = 0;
+		
+					// sandbox_meta->remaining_exec -= new_message.adjustment;
+					assert(sandbox_meta->remaining_exec == new_message.remaining_exec);
+				}
+				// assert(sandbox_meta->remaining_exec == 0);
+
+				if (sandbox_meta->trs_job_node) tenant_reduce_guaranteed_job_demand(tenant, sandbox_meta->remaining_exec, sandbox_meta);
+
+				// if (sandbox_meta->global_queue_type == 2 && sandbox_meta->state == SANDBOX_RETURNED) sandbox_meta->total_be_exec_cycles += new_message.adjustment;
+				if (sandbox_meta->total_be_exec_cycles > 0) tenant_force_add_job_as_best(tenant, now, sandbox_meta->total_be_exec_cycles);
+
+				sandbox_meta->terminated = true;
+				break;
+			}
+			case MESSAGE_CFW_REDUCE_DEMAND: {
+				assert(new_message.adjustment > 0);
+				if (sandbox_meta->global_queue_type == 2) {
+					sandbox_meta->total_be_exec_cycles += new_message.adjustment;
+					// tenant_force_add_job_as_best(tenant, now, sandbox_meta->total_be_exec_cycles);
+				}
+
+				if (sandbox_meta->terminated) break;
+				
+				assert(sandbox_meta->demand_node);
+				assert(sandbox_meta->worker_id_virt >= 0);
+				void *global_dbf = global_virt_worker_dbfs[sandbox_meta->worker_id_virt];
+				dbf_list_reduce_demand(sandbox_meta, new_message.adjustment, false);
+	
+				sandbox_meta->remaining_exec -= new_message.adjustment;
+				assert(sandbox_meta->remaining_exec == new_message.remaining_exec);
+
+				if (sandbox_meta->remaining_exec == 0 && sandbox_meta->extra_slack < runtime_quantum) {
+					dbf_list_reduce_demand(sandbox_meta, sandbox_meta->extra_slack, true);
+					sandbox_meta->extra_slack = 0;
+					sandbox_meta->demand_node = NULL;
+				}
+				break;
+			}
+			case MESSAGE_CFW_EXTRA_DEMAND_REQUEST: {
+				assert(USING_TRY_LOCAL_EXTRA);
+				assert(new_message.exceeded_estimation);
+				assert(new_message.adjustment == runtime_quantum);
+
+				if (sandbox_meta->terminated) break;
+
+				assert(sandbox_meta->remaining_exec == 0);
+				if (absolute_deadline <= now) {
+					assert(sandbox_meta->error_code == 0);
+					if (sandbox_meta->extra_slack > 0) {
+						assert(sandbox_meta->demand_node);
+						dbf_list_reduce_demand(sandbox_meta, sandbox_meta->extra_slack, true);
+						sandbox_meta->demand_node = NULL;
+						// sandbox_meta->extra_slack = 0;
+					}
+					assert(sandbox_meta->demand_node == NULL);
+					sandbox_meta->error_code = 4081;
+					sandbox_meta->terminated = true;
+					break;
+				}
+
+				if (sandbox_meta->global_queue_type == 1) {
+					assert(sandbox_meta->trs_job_node);
+					assert(sandbox_meta->trs_job_node->sandbox_meta);
+					sandbox_meta->trs_job_node->sandbox_meta = NULL;
+					sandbox_meta->trs_job_node = NULL;
+				}
+
+				int 	 return_code   = 0;
+				int 	 worker_id_v   = -1;
+				uint64_t work_admitted = 0;
+				
+				if (sandbox_meta->extra_slack >= runtime_quantum) {
+					bool tenant_can_admit = tenant_try_add_job_as_guaranteed(tenant, now, runtime_quantum, sandbox_meta);
+					return_code = tenant_can_admit ? 1 : 2;
+
+					worker_id_v = sandbox_meta->worker_id_virt;
+					work_admitted = 1;
+
+					sandbox_meta->extra_slack -= runtime_quantum;
+				} else {
+					assert(sandbox_meta->demand_node == NULL);
+					assert(sandbox_meta->extra_slack == 0);
+					work_admitted = traffic_control_decide(sandbox_meta, now, runtime_quantum, &return_code, &worker_id_v);
+				}
+
+				if (work_admitted == 0) {
+					// debuglog("No global supply left");
+					assert(return_code == 4295 || return_code == 4296);
+					assert(sandbox_meta->demand_node == NULL);
+					assert(sandbox_meta->extra_slack == 0);
+					assert(sandbox_meta->error_code == 0);
+					sandbox_meta->error_code = return_code;
+					sandbox_meta->terminated = true;
+					break;
+				}
+
+				assert(worker_id_v >= 0);
+				sandbox_meta->remaining_exec = runtime_quantum;
+				sandbox_meta->worker_id_virt = worker_id_v;
+
+				// TODO: Fix the BE budget calculation for when promote/demote happens
+				if (sandbox_meta->global_queue_type == 2 && return_code == 1) {
+					assert(sandbox_meta->pq_idx_in_tenant_queue >= 1);
+					priority_queue_delete_by_idx_nolock(tenant->local_sandbox_metas,
+											sandbox_meta,sandbox_meta->pq_idx_in_tenant_queue);
+					sandbox_meta->tenant_queue = NULL;
+					// printf("promote!\n");
+				} else if (sandbox_meta->global_queue_type == 1 && return_code == 2) {
+					assert(sandbox_meta->pq_idx_in_tenant_queue == 0);
+					if(unlikely(priority_queue_enqueue_nolock(tenant->local_sandbox_metas, sandbox_meta))){
+						panic("Failed to add sandbox_meta to tenant metadata queue");
+					}
+					sandbox_meta->tenant_queue = tenant->local_sandbox_metas;
+					// printf("demote!\n");
+				}
+				sandbox_meta->global_queue_type = return_code;
+				break;
+			}
+			case MESSAGE_CFW_WRITEBACK_PREEMPTION: {
+				assert(USING_WRITEBACK_FOR_PREEMPTION);
+				assert(USING_LOCAL_RUNQUEUE == false);
+				assert (sandbox_meta->state == SANDBOX_PREEMPTED);
+
+				struct sandbox  *preempted_sandbox = new_message.sandbox;
+				assert(preempted_sandbox);
+				assert(preempted_sandbox == sandbox_meta->sandbox_shadow);
+				assert(preempted_sandbox->sandbox_meta == sandbox_meta);
+				assert(preempted_sandbox->id == new_message.sandbox_id);
+				assert(preempted_sandbox->state == SANDBOX_PREEMPTED);
+				assert(preempted_sandbox->writeback_preemption_in_progress);
+				assert(preempted_sandbox->absolute_deadline == absolute_deadline);
+				assert(preempted_sandbox->response_code == 0);
+				assert(preempted_sandbox->remaining_exec > 0);
+
+				if (sandbox_meta->terminated) {
+					assert(sandbox_meta->error_code > 0);
+					assert(preempted_sandbox->response_code == 0);
+					// preempted_sandbox->response_code = sandbox_meta->error_code;
+					// break;
+					// printf("terminated - %s\n", tenant->name);
+				} else if (absolute_deadline < now + (!preempted_sandbox->exceeded_estimation ? preempted_sandbox->remaining_exec : 0)) {
+					// printf("missed - %s\n", tenant->name);
+		/*		// if (absolute_deadline < now + preempted_sandbox->remaining_execution_original) {
+					assert(sandbox_meta->terminated == false);
+					assert(sandbox_meta->remaining_exec == preempted_sandbox->remaining_exec);
+
+					assert(sandbox_meta->worker_id_virt >= 0);
+					void *global_dbf = global_virt_worker_dbfs[sandbox_meta->worker_id_virt];
+					dbf_try_update_demand(global_dbf, allocation_timestamp,
+										route->relative_deadline,
+										absolute_deadline, sandbox_meta->remaining_exec,
+										DBF_DELETE_EXISTING_DEMAND, NULL, NULL);
+
+					sandbox_meta->remaining_exec -= new_message.adjustment;
+					assert(sandbox_meta->remaining_exec == new_message.remaining_exec);
+
+					if (sandbox_meta->trs_job_node) {
+						assert(sandbox_meta->global_queue_type == 1);
+						tenant_update_job_node(tenant, sandbox_meta->remaining_exec, TRS_REDUCE_EXISTING_DEMAND, sandbox_meta);
+					}  
+					
+					if (sandbox_meta->global_queue_type == 2) {
+						assert(tenant->trs.best_effort_cycles >= sandbox_meta->remaining_exec);
+						tenant->trs.best_effort_cycles -= sandbox_meta->remaining_exec;
+					}
+
+					sandbox_meta->terminated = true;
+					assert(preempted_sandbox->response_code == 0);
+					preempted_sandbox->response_code = 4082;
+					break; */
+					assert(sandbox_meta->error_code == 0);
+					if (sandbox_meta->remaining_exec + sandbox_meta->extra_slack > 0) {
+						assert(sandbox_meta->demand_node);
+						dbf_list_reduce_demand(sandbox_meta, sandbox_meta->remaining_exec + sandbox_meta->extra_slack, true);
+						sandbox_meta->demand_node = NULL;
+						// sandbox_meta->remaining_exec = 0;
+						// sandbox_meta->extra_slack = 0;
+					}
+					assert(sandbox_meta->demand_node == NULL);
+					sandbox_meta->error_code = 4082;
+					sandbox_meta->terminated = true;
+				}
+				
+				assert(sandbox_meta->terminated || sandbox_meta->remaining_exec == preempted_sandbox->remaining_exec);
+
+				if (unlikely(global_request_scheduler_add(preempted_sandbox) == NULL)) {
+					// TODO: REDUCE DBF, for now just panic!
+					panic("Failed to add the preempted_sandbox to global queue\n");
+				}
+				preempted_sandbox->writeback_preemption_in_progress = false;
+				
+				break;
+			}
+			case MESSAGE_CFW_WRITEBACK_OVERSHOOT: {
+				assert(USING_WRITEBACK_FOR_OVERSHOOT);
+				assert (sandbox_meta->state == SANDBOX_PREEMPTED);
+
+				struct sandbox  *preempted_sandbox = new_message.sandbox;
+				assert(preempted_sandbox);
+				assert(preempted_sandbox == sandbox_meta->sandbox_shadow);
+				assert(preempted_sandbox->sandbox_meta == sandbox_meta);
+				assert(preempted_sandbox->id == new_message.sandbox_id);
+				assert(preempted_sandbox->sandbox_meta == sandbox_meta);
+				assert(preempted_sandbox->state == SANDBOX_PREEMPTED);
+				assert(preempted_sandbox->writeback_overshoot_in_progress);
+				assert(preempted_sandbox->absolute_deadline == absolute_deadline);
+				assert(preempted_sandbox->response_code == 0);
+				assert(preempted_sandbox->remaining_exec == 0);
+				assert(new_message.remaining_exec == 0);
+				assert(sandbox_meta->remaining_exec == 0);
+				assert(new_message.adjustment == runtime_quantum);		
+				
+				if (sandbox_meta->terminated) {
+					assert(sandbox_meta->error_code > 0);
+					preempted_sandbox->response_code = sandbox_meta->error_code;
+					break;
+				}
+
+				if (absolute_deadline <= now) {
+					sandbox_meta->terminated = true;
+					preempted_sandbox->response_code = 4082;
+					break;
+				}
+
+				if (sandbox_meta->global_queue_type == 1) {
+					assert(sandbox_meta->trs_job_node);
+					assert(sandbox_meta->trs_job_node->sandbox_meta);
+					sandbox_meta->trs_job_node->sandbox_meta = NULL;
+					sandbox_meta->trs_job_node = NULL;
+				}
+				
+				int 	 return_code   = 0;
+				int 	 worker_id_v   = -1;
+				uint64_t work_admitted = 0;
+				
+				work_admitted = traffic_control_decide(sandbox_meta, now, runtime_quantum,
+									&return_code, &worker_id_v);
+
+				if (work_admitted == 0) {
+					// debuglog("No global supply left");
+					assert(return_code == 4290 || return_code == 4291);
+					preempted_sandbox->response_code = return_code + 5;
+					sandbox_meta->terminated = true;
+					break;
+				}
+
+				assert(worker_id_v >= 0);
+				sandbox_meta->remaining_exec = runtime_quantum;
+				sandbox_meta->worker_id_virt = worker_id_v;
+				if (sandbox_meta->global_queue_type == 2 && return_code == 1) {
+					assert(sandbox_meta->pq_idx_in_tenant_queue >= 1);
+					priority_queue_delete_by_idx_nolock(tenant->local_sandbox_metas,
+											sandbox_meta,sandbox_meta->pq_idx_in_tenant_queue);
+				} else if (sandbox_meta->global_queue_type == 1 && return_code == 2) {
+					assert(sandbox_meta->pq_idx_in_tenant_queue == 0);
+					if(unlikely(priority_queue_enqueue_nolock(tenant->local_sandbox_metas, sandbox_meta))){
+						panic("Failed to add sandbox_meta to tenant metadata queue");
+					}
+				}
+				sandbox_meta->global_queue_type = return_code;
+
+				preempted_sandbox->remaining_exec = runtime_quantum;
+				preempted_sandbox->writeback_overshoot_in_progress = false;
+
+				if (unlikely(global_request_scheduler_add(preempted_sandbox) == NULL)){
+					// TODO: REDUCE DBF, for now just panic!
+					panic("Failed to add the preempted_sandbox to global queue\n");
+				}
+				break;
+			}
+			default:
+				panic("Unknown message type received by the listener!");
+				break;
+			} // end switch 1
+
+			if (sandbox_meta->terminated == false) continue;
+
+			assert(sandbox_meta->demand_node == NULL);
+			
+			if (sandbox_meta->pq_idx_in_tenant_queue) {
+				assert(sandbox_meta->global_queue_type == 2);
+				assert(sandbox_meta->tenant_queue);
+				priority_queue_delete_by_idx_nolock(sandbox_meta->tenant_queue,
+										sandbox_meta,sandbox_meta->pq_idx_in_tenant_queue);
+				sandbox_meta->tenant_queue = NULL;
+			}
+
+			switch (new_message.message_type) {
+			case MESSAGE_CFW_EXTRA_DEMAND_REQUEST:
+				if (sandbox_refs[new_message.sandbox_id % RUNTIME_MAX_ALIVE_SANDBOXES]){
+					new_message.message_type = MESSAGE_CTW_SHED_CURRENT_JOB;
+					if (!ck_ring_enqueue_spsc_message(&ctw->worker_ring, ctw->worker_ring_buffer, &new_message)) {
+						panic("Ring buffer was full and enqueue has failed!")
+					}
+					pthread_kill(runtime_worker_threads[new_message.sender_worker_idx], SIGALRM);
+				} else {
+					// printf("already dead\n");
+				}
+				break;
+			case MESSAGE_CFW_REDUCE_DEMAND:
+				break;
+			case MESSAGE_CFW_DELETE_SANDBOX:
+				assert (sandbox_meta->state == SANDBOX_RETURNED || sandbox_meta->state == SANDBOX_ERROR);
+				free(sandbox_meta);
+				break;
+			case MESSAGE_CFW_WRITEBACK_PREEMPTION:
+			case MESSAGE_CFW_WRITEBACK_OVERSHOOT:
+				// assert(preempted_sandbox);
+				// assert(preempted_sandbox->id == sandbox_meta->id);
+				// assert(preempted_sandbox->state == SANDBOX_PREEMPTED);
+
+				// sandbox_set_as_error(preempted_sandbox, SANDBOX_PREEMPTED);
+				// sandbox_free(preempted_sandbox);
+				// free(sandbox_meta);
+				break;
+			default:
+				panic ("Unknown message type!");
+				break;
+			} // end switch 2
+			// memset(&new_message, 0, sizeof(new_message));
+		} // end while
+	} // end for
+
+#endif
+}
+
 static void
 panic_on_epoll_error(struct epoll_event *evt)
 {
@@ -239,56 +655,201 @@ static void
 on_client_request_received(struct http_session *session)
 {
 	assert(session->state == HTTP_SESSION_RECEIVED_REQUEST);
-	session->request_downloaded_timestamp = __getcycles();
-	struct route *route                   = session->route;
-	uint64_t      estimated_execution     = route->execution_histogram.estimated_execution;
-	uint64_t      work_admitted           = 1;
+	const uint64_t now = __getcycles();
+	session->request_downloaded_timestamp = now;
+
+	struct tenant *tenant = session->tenant;
+	struct route  *route  = session->route;
+	http_route_total_increment_request(&session->route->metrics);
+
+	// uint64_t estimated_execution = route->execution_histogram.estimated_execution; // By defaulat this is half of deadline
+	// uint64_t work_admitted       = 1;
+
+	// struct route *route = http_router_match_route(&tenant->router, session->http_request.full_url);
+	// if (route == NULL) {
+	// 	debuglog("Route: %s did not match any routes\n", session->http_request.full_url);
+	// 	session->state = HTTP_SESSION_EXECUTION_COMPLETE;
+	// 	http_session_set_response_header(session, 404);
+	// 	on_client_response_header_sending(session);
+	// 	return;
+	// }
+
+	// session->route = route;
+	// http_route_total_increment_request(&session->route->metrics);
+
+#if defined TRAFFIC_CONTROL
+	/*
+	 * Admin control.
+	 * If client sends a request to the route "/main", server prints all the DBF data.
+	 * If client sends a request to the route "/terminator", server does cleanup and terminates.
+	 */
+	if (tenant->tcp_server.port == 55555) {
+		if (strcmp(session->http_request.full_url, "/terminator") == 0) {
+			printf("Terminating SLEdge now!\n");
+			tenant_database_print_reservations();
+			printf("\nGLOBAL DBF DEMANDS:\n");
+			const int N_VIRT_WORKERS_DBF = USING_AGGREGATED_GLOBAL_DBF ? 1 : runtime_worker_threads_count;
+			for (int i = 0; i < N_VIRT_WORKERS_DBF; i++)
+			{
+				printf("GL Worker #%d\n", i);
+				dbf_print(global_virt_worker_dbfs[i], now);
+			}
+			
+			// dbf_print(global_dbf, now);
+			// printf("\nGLOBAL GUAR DBF DEMANDS:\n");
+			printf("\nWorker #0 DBF DEMANDS:\n");
+			dbf_print(global_worker_dbf, now);
+			
+			session->state = HTTP_SESSION_EXECUTION_COMPLETE;
+			http_session_set_response_header(session, 500);
+			on_client_response_header_sending(session);
+			runtime_cleanup();
+			exit(0);
+		}
+
+		if (strcmp(session->http_request.full_url, "/admin") == 0) {
+			printf("Hello from Admin!\n");
+			tenant_database_print_reservations();
+			printf("\nGLOBAL DBF DEMANDS:\n");
+			const int N_VIRT_WORKERS_DBF = USING_AGGREGATED_GLOBAL_DBF ? 1 : runtime_worker_threads_count;
+			for (int i = 0; i < N_VIRT_WORKERS_DBF; i++)
+			{
+				printf("GL Worker #%d\n", i);
+				dbf_print(global_virt_worker_dbfs[i], now);
+			}
+			
+			// dbf_print(global_dbf, now);
+			// printf("\nGLOBAL GUAR DBF DEMANDS:\n");
+			printf("\nWorker #0 DBF DEMANDS:\n");
+			dbf_print(global_worker_dbf, now);
+			
+			session->state = HTTP_SESSION_EXECUTION_COMPLETE;
+			http_session_set_response_header(session, 500);
+			on_client_response_header_sending(session);
+			return;
+		}
+	}
+#endif
+
+	// const uint64_t sandbox_alloc_timestamp = __getcycles();
+	const uint64_t absolute_deadline = now + route->relative_deadline;
+	uint64_t estimated_execution     = route->execution_histogram.estimated_execution; // TODO: By default half of deadline
+	uint64_t       work_admitted     = 1;
+	int            return_code       = 0;
+	int            worker_id_v       = -1;
 
 #ifdef EXECUTION_REGRESSION
 	estimated_execution = get_regression_prediction(session);
 #endif
 
-#ifdef ADMISSIONS_CONTROL
-	/*
-	 * Perform admissions control.
-	 * If 0, workload was rejected, so close with 429 "Too Many Requests" and continue
-	 */
-	uint64_t admissions_estimate = admissions_control_calculate_estimate(estimated_execution,
+/*
+ * Perform admissions control.
+ * If 0, workload was rejected, so close with 429 "Too Many Requests" and continue
+ */
+#if defined ADMISSIONS_CONTROL
+	const uint64_t admissions_estimate = admissions_control_calculate_estimate(estimated_execution,
 	                                                                     route->relative_deadline);
 	work_admitted                = admissions_control_decide(admissions_estimate);
+#elif defined TRAFFIC_CONTROL
+	if (global_sandbox_meta == NULL) global_sandbox_meta = malloc(sizeof(struct sandbox_metadata));
+
+	assert(global_sandbox_meta);
+	global_sandbox_meta->tenant = tenant;
+	global_sandbox_meta->route = route;
+	global_sandbox_meta->tenant_queue = NULL;
+	global_sandbox_meta->sandbox_shadow = NULL;
+	global_sandbox_meta->global_queue_type = 0;
+	global_sandbox_meta->pq_idx_in_tenant_queue = 0;
+	global_sandbox_meta->error_code = 0;
+	global_sandbox_meta->exceeded_estimation = false;
+	global_sandbox_meta->terminated = false;
+	global_sandbox_meta->demand_node = NULL;
+	global_sandbox_meta->trs_job_node = NULL;
+	global_sandbox_meta->extra_slack = 0;
+	global_sandbox_meta->total_be_exec_cycles = 0;
+	global_sandbox_meta->owned_worker_idx = -2;
+
+	global_sandbox_meta->allocation_timestamp = now;
+	global_sandbox_meta->absolute_deadline = absolute_deadline;
+	global_sandbox_meta->remaining_exec  = estimated_execution;
+	
+	work_admitted = traffic_control_decide(global_sandbox_meta, now, estimated_execution,
+	                                       &return_code, &worker_id_v);
+	assert(work_admitted == 0 || worker_id_v >= 0);
+	assert(work_admitted == 0 || return_code == 1 || return_code == 2);
+#endif
+
 	if (work_admitted == 0) {
+		assert(worker_id_v < 0);
 		session->state = HTTP_SESSION_EXECUTION_COMPLETE;
 		http_session_set_response_header(session, 429);
 		on_client_response_header_sending(session);
+		sandbox_perf_log_print_denied_entry(tenant, route, return_code);
 		return;
 	}
-#endif
 
 	/* Allocate a Sandbox */
 	session->state          = HTTP_SESSION_EXECUTING;
-	struct sandbox *sandbox = sandbox_alloc(route->module, session, route, session->tenant, work_admitted);
+	struct sandbox *sandbox = sandbox_alloc(route->module, session, work_admitted, now);
 	if (unlikely(sandbox == NULL)) {
+		// TODO: REDUCE DEMAND!!!
 		debuglog("Failed to allocate sandbox\n");
 		session->state = HTTP_SESSION_EXECUTION_COMPLETE;
 		http_session_set_response_header(session, 500);
 		on_client_response_header_sending(session);
+		sandbox_perf_log_print_denied_entry(tenant, route, 5000);
 		return;
 	}
 
 	sandbox->remaining_exec = estimated_execution;
+	
+#if defined TRAFFIC_CONTROL	
+	sandbox->global_queue_type = return_code;
+	sandbox->sandbox_meta = global_sandbox_meta;
+
+	if(extra_execution_slack_p > 0) {
+		const uint64_t hack = estimated_execution*extra_execution_slack_p/100;
+		dbf_list_force_add_extra_slack(global_virt_worker_dbfs[worker_id_v], global_sandbox_meta, hack);
+		global_sandbox_meta->extra_slack = hack;
+	}
+
+	global_sandbox_meta->sandbox_shadow       = sandbox;
+	global_sandbox_meta->id                   = sandbox->id;
+	global_sandbox_meta->state                = sandbox->state;
+
+	global_sandbox_meta->worker_id_virt = worker_id_v;
+	global_sandbox_meta->global_queue_type = return_code;
+
+	if (global_sandbox_meta->global_queue_type == 2) {
+		if(unlikely(priority_queue_enqueue_nolock(tenant->global_sandbox_metas, global_sandbox_meta))){
+			panic("Failed to add sandbox_meta to tenant metadata queue");
+		}
+		global_sandbox_meta->tenant_queue = tenant->global_sandbox_metas;
+	}
+#endif
 
 	/* If the global request scheduler is full, return a 429 to the client */
 	if (unlikely(global_request_scheduler_add(sandbox) == NULL)) {
-		// debuglog("Failed to add sandbox to global queue\n");
+		// debuglog("Failed to add a %s sandbox to global queue\n", sandbox->tenant->name);
+		/////////////////////////////////// TODO ???
 		sandbox->response_code = 4290;
-		sandbox->state         = SANDBOX_ERROR;
-		sandbox_perf_log_print_entry(sandbox);
-		sandbox->http = NULL;
+		// sandbox->state         = SANDBOX_ERROR;
+		// sandbox_perf_log_print_entry(sandbox);
+		// sandbox->http = NULL;
+
+		sandbox->timestamp_of.dispatched = now;
+		// TODO: REDUCE DEMAND!!!
+		sandbox_set_as_error(sandbox, SANDBOX_INITIALIZED);
+		free(sandbox->sandbox_meta);
 		sandbox_free(sandbox);
-		session->state = HTTP_SESSION_EXECUTION_COMPLETE;
-		http_session_set_response_header(session, 429);
-		on_client_response_header_sending(session);
+		return;
+		// session->state = HTTP_SESSION_EXECUTION_COMPLETE;
+		// http_session_set_response_header(session, 429);
+		// on_client_response_header_sending(session);
+		// sandbox_perf_log_print_denied_entry(tenant, route, 999);
 	}
+
+	global_sandbox_meta = NULL;
 }
 
 static void
@@ -437,20 +998,30 @@ listener_thread_main(void *dummy)
 	listener_thread_register_metrics_server();
 
 	/* Set my priority */
-	// runtime_set_pthread_prio(pthread_self(), 2);
+	// runtime_set_pthread_prio(pthread_self(), 2); // TODO: what to do with this?
 	pthread_setschedprio(pthread_self(), -20);
 
+#ifdef TRAFFIC_CONTROL
+	const int epoll_timeout = 0;
+#else
+	const int epoll_timeout = -1;
+#endif
+
 	while (true) {
-		/* Block indefinitely on the epoll file descriptor, waiting on up to a max number of events */
+#ifdef TRAFFIC_CONTROL
+		tenant_database_replenish_all();
+		check_messages_from_workers();
+#endif		
+		/* If -1, Block indefinitely on the epoll file descriptor, waiting on up to a max number of events */
 		int descriptor_count = epoll_wait(listener_thread_epoll_file_descriptor, epoll_events,
-		                                  RUNTIME_MAX_EPOLL_EVENTS, -1);
+		                                  RUNTIME_MAX_EPOLL_EVENTS, epoll_timeout);
+
+		if (descriptor_count == 0) continue;
 		if (descriptor_count < 0) {
 			if (errno == EINTR) continue;
 
 			panic("epoll_wait: %s", strerror(errno));
 		}
-
-		/* Assumption: Because epoll_wait is set to not timeout, we should always have descriptors here */
 		assert(descriptor_count > 0);
 
 		for (int i = 0; i < descriptor_count; i++) {
@@ -460,6 +1031,8 @@ listener_thread_main(void *dummy)
 
 			switch (tag) {
 			case EPOLL_TAG_TENANT_SERVER_SOCKET:
+				// tenant_database_replenish_all();
+				// check_messages_from_workers();
 				on_tenant_socket_epoll_event(&epoll_events[i]);
 				break;
 			case EPOLL_TAG_HTTP_SESSION_CLIENT_SOCKET:

runtime/src/local_cleanup_queue.c

@@ -5,7 +5,7 @@
 
 /* Must be the same alignment as sandbox structs because of how the ps_list macros work */
 thread_local static struct ps_list_head local_cleanup_queue PAGE_ALIGNED;
-
+thread_local size_t local_cleanup_queue_size = 0;
 
 void
 local_cleanup_queue_initialize()
@@ -29,6 +29,7 @@ local_cleanup_queue_add(struct sandbox *sandbox)
 	assert(sandbox);
 	assert(ps_list_singleton_d(sandbox));
 	ps_list_head_append_d(&local_cleanup_queue, sandbox);
+	local_cleanup_queue_size++;
 	assert(!local_cleanup_queue_is_empty());
 }
 
@@ -42,10 +43,13 @@ local_cleanup_queue_free()
 {
 	struct sandbox *sandbox_iterator = NULL;
 	struct sandbox *buffer           = NULL;
+	// if (local_cleanup_queue_size > 4) debuglog("Cleanup Queue Size: %lu", local_cleanup_queue_size);
 
 	ps_list_foreach_del_d(&local_cleanup_queue, sandbox_iterator, buffer)
 	{
+		assert(local_cleanup_queue_size > 0);
 		ps_list_rem_d(sandbox_iterator);
 		sandbox_free(sandbox_iterator);
+		local_cleanup_queue_size--;
 	}
 }

runtime/src/local_runqueue_minheap.c

@@ -14,6 +14,8 @@
 
 thread_local static struct priority_queue *local_runqueue_minheap;
 
+thread_local static int max_local_runqueue_len = 0; //////////
+
 /**
  * Checks if the run queue is empty
  * @returns true if empty. false otherwise
@@ -40,6 +42,11 @@ local_runqueue_minheap_add(struct sandbox *sandbox)
 		return_code            = priority_queue_enqueue_nolock(local_runqueue_minheap, sandbox);
 		if (unlikely(return_code == -ENOSPC)) panic("Thread Runqueue is full!\n");
 	}
+
+	if(priority_queue_length_nolock(local_runqueue_minheap) > max_local_runqueue_len) {
+		max_local_runqueue_len = priority_queue_length_nolock(local_runqueue_minheap);
+		debuglog("Local MAX Queue Length: %u", max_local_runqueue_len);
+	}
 }
 
 /**

runtime/src/local_runqueue_mtdbf.c

@@ -0,0 +1,124 @@
+#include <stdint.h>
+#include <threads.h>
+
+#include "arch/context.h"
+#include "current_sandbox.h"
+#include "debuglog.h"
+#include "global_request_scheduler.h"
+#include "local_runqueue.h"
+#include "local_runqueue_mtdbf.h"
+#include "panic.h"
+#include "priority_queue.h"
+#include "sandbox_functions.h"
+#include "runtime.h"
+#include "dbf.h"
+
+thread_local  struct priority_queue *local_runqueue_mtdbf;
+// thread_local  struct priority_queue *local_default_queue;
+
+thread_local static int max_local_runqueue_len = 0; //////////
+
+/**
+ * Checks if the run queue is empty
+ * @returns true if empty. false otherwise
+ */
+bool
+local_runqueue_mtdbf_is_empty()
+{
+	return priority_queue_length_nolock(local_runqueue_mtdbf) == 0;
+}
+
+/**
+ * Adds a sandbox to the run queue
+ * @param sandbox
+ * @returns pointer to sandbox added
+ */
+void
+local_runqueue_mtdbf_add(struct sandbox *sandbox)
+{
+	assert(sandbox != NULL);
+
+	int rc = priority_queue_enqueue_nolock(local_runqueue_mtdbf, sandbox);
+	if (unlikely(rc == -ENOSPC)) {
+		struct priority_queue *temp = priority_queue_grow_nolock(local_runqueue_mtdbf);
+		if (unlikely(temp == NULL)) panic("Failed to grow local runqueue\n");
+		local_runqueue_mtdbf = temp;
+		rc                   = priority_queue_enqueue_nolock(local_runqueue_mtdbf, sandbox);
+		if (unlikely(rc == -ENOSPC)) panic("Thread Runqueue is full!\n");
+	}
+
+	// if (sandbox->global_queue_type == 2) {
+	// 	rc = priority_queue_enqueue_nolock(local_default_queue, sandbox);
+	// 	assert(rc == 0);
+	// }
+
+	sandbox->owned_worker_idx = worker_thread_idx;
+
+	if(priority_queue_length_nolock(local_runqueue_mtdbf) > max_local_runqueue_len) {
+		max_local_runqueue_len = priority_queue_length_nolock(local_runqueue_mtdbf);
+		debuglog("Local MAX Queue Length: %u", max_local_runqueue_len);
+	}
+}
+
+/**
+ * Deletes a sandbox from the runqueue
+ * @param sandbox to delete
+ */
+static void
+local_runqueue_mtdbf_delete(struct sandbox *sandbox)
+{
+	assert(sandbox != NULL);
+
+	priority_queue_delete_by_idx_nolock(local_runqueue_mtdbf, sandbox, sandbox->pq_idx_in_runqueue);
+	sandbox->owned_worker_idx = -2;
+	// if (sandbox->pq_idx_in_default_queue >= 1) {
+	// 	assert(sandbox->global_queue_type == 2 );
+	// 	priority_queue_delete_by_idx_nolock(local_default_queue, sandbox, sandbox->pq_idx_in_default_queue);
+	// }
+}
+
+/**
+ * This function determines the next sandbox to run.
+ * This is the head of the runqueue
+ *
+ * Execute the sandbox at the head of the thread local runqueue
+ * @return the sandbox to execute or NULL if none are available
+ */
+struct sandbox *
+local_runqueue_mtdbf_get_next()
+{
+	/* Get the deadline of the sandbox at the head of the local request queue */
+	struct sandbox *next = NULL;
+	int             rc   = priority_queue_top_nolock(local_runqueue_mtdbf, (void **)&next);
+
+	if (rc == -ENOENT) return NULL;
+
+	return next;
+}
+
+// static inline void
+// sandbox_update_pq_idx_in_default_queue(void *element, size_t idx)
+// {
+// 	assert(element);
+// 	struct sandbox *sandbox     = (struct sandbox *)element;
+// 	sandbox->pq_idx_in_default_queue = idx;
+// }
+
+/**
+ * Registers the PS variant with the polymorphic interface
+ */
+void
+local_runqueue_mtdbf_initialize()
+{
+	/* Initialize local state */
+	local_runqueue_mtdbf = priority_queue_initialize_new(RUNTIME_RUNQUEUE_SIZE, false, sandbox_get_priority_global, NULL,
+	                                                     sandbox_update_pq_idx_in_runqueue);
+
+	/* Register Function Pointers for Abstract Scheduling API */
+	struct local_runqueue_config config = { .add_fn      = local_runqueue_mtdbf_add,
+		                                .is_empty_fn = local_runqueue_mtdbf_is_empty,
+		                                .delete_fn   = local_runqueue_mtdbf_delete,
+		                                .get_next_fn = local_runqueue_mtdbf_get_next };
+
+	local_runqueue_initialize(&config);
+}

runtime/src/main.c

@@ -31,7 +31,7 @@
 
 /* Conditionally used by debuglog when NDEBUG is not set */
 int32_t  debuglog_file_descriptor        = -1;
-uint32_t runtime_first_worker_processor  = 1;
+uint32_t runtime_first_worker_processor  = 2;
 uint32_t runtime_processor_speed_MHz     = 0;
 uint32_t runtime_total_online_processors = 0;
 uint32_t runtime_worker_threads_count    = 0;
@@ -41,8 +41,11 @@ enum RUNTIME_SIGALRM_HANDLER runtime_sigalrm_handler = RUNTIME_SIGALRM_HANDLER_B
 bool     runtime_preemption_enabled            = true;
 bool     runtime_worker_spinloop_pause_enabled = false;
 uint32_t runtime_quantum_us                    = 1000; /* 1ms */
+uint64_t runtime_quantum;                   /* cycles */
 uint64_t runtime_boot_timestamp;
+uint64_t runtime_max_deadline = 0L;
 pid_t    runtime_pid = 0;
+uint16_t extra_execution_slack_p = 0; /* percentile */
 
 /**
  * Returns instructions on use of CLI if used incorrectly
@@ -69,7 +72,7 @@ runtime_allocate_available_cores()
 
 	/* If more than two cores are available, leave core 0 free to run OS tasks */
 	if (runtime_total_online_processors > 2) {
-		runtime_first_worker_processor = 2;
+		// runtime_first_worker_processor = 2;
 		max_possible_workers           = runtime_total_online_processors - 2;
 	} else if (runtime_total_online_processors == 2) {
 		runtime_first_worker_processor = 1;
@@ -105,8 +108,12 @@ runtime_allocate_available_cores()
 static inline void
 runtime_get_processor_speed_MHz(void)
 {
+	// runtime_processor_speed_MHz = 2600; // nworkers=1
+	// // runtime_processor_speed_MHz = 2893; // nworkers > 4
+	// return; ////////////////////////// temp
+
 	char *proc_mhz_raw = getenv("SLEDGE_PROC_MHZ");
-	FILE *cmd          = NULL;
+	// FILE *cmd          = NULL;
 
 	if (proc_mhz_raw != NULL) {
 		/* The case with manual override for the CPU freq */
@@ -121,7 +128,7 @@ runtime_get_processor_speed_MHz(void)
 					awk '{ total += $4; count++ } END { print total/count }'",
 		        runtime_first_worker_processor + 1,
 		        runtime_first_worker_processor + runtime_worker_threads_count);
-		cmd = popen(command, "r");
+		FILE *cmd = popen(command, "r");
 		if (unlikely(cmd == NULL)) goto err;
 
 		char   buff[16];
@@ -140,6 +147,7 @@ runtime_get_processor_speed_MHz(void)
 	pretty_print_key_value("Worker CPU Freq", "%u MHz\n", runtime_processor_speed_MHz);
 
 done:
+	// pclose(cmd);
 	return;
 err:
 	goto done;
@@ -223,7 +231,7 @@ runtime_configure()
 	if (strcmp(sigalrm_policy, "BROADCAST") == 0) {
 		runtime_sigalrm_handler = RUNTIME_SIGALRM_HANDLER_BROADCAST;
 	} else if (strcmp(sigalrm_policy, "TRIAGED") == 0) {
-		if (unlikely(scheduler != SCHEDULER_EDF)) panic("triaged sigalrm handlers are only valid with EDF\n");
+		// if (unlikely(scheduler != SCHEDULER_EDF)) panic("triaged sigalrm handlers are only valid with EDF\n");
 		runtime_sigalrm_handler = RUNTIME_SIGALRM_HANDLER_TRIAGED;
 	} else {
 		panic("Invalid sigalrm policy: %s. Must be {BROADCAST|TRIAGED}\n", sigalrm_policy);
@@ -245,7 +253,17 @@ runtime_configure()
 			panic("SLEDGE_QUANTUM_US must be less than 999999 ms, saw %ld\n", quantum);
 		runtime_quantum_us = (uint32_t)quantum;
 	}
-	pretty_print_key_value("Quantum", "%u us\n", runtime_quantum_us);
+	// runtime_quantum = runtime_quantum_us * runtime_processor_speed_MHz;
+	// pretty_print_key_value("Quantum", "%u us = %lu cycles\n", runtime_quantum_us, runtime_quantum);
+
+	/* Extra execution slack percentile */
+	char *extra_slack_raw = getenv("EXTRA_EXEC_PERCENTILE");
+	if (extra_slack_raw != NULL) {
+		int extra_slack = atoi(extra_slack_raw);
+		if (unlikely(extra_slack < 0 || extra_slack > 50)) panic("EXTRA_EXEC_PERCENTILE must be between [0-50], saw %d\n", extra_slack);
+		extra_execution_slack_p = (uint16_t)extra_slack;
+	}
+	pretty_print_key_value("Extra Exec Slack", "%u%%\n", extra_execution_slack_p);
 
 	sandbox_perf_log_init();
 	http_session_perf_log_init();
@@ -391,6 +409,12 @@ log_compiletime_config()
 #else
 	pretty_print_key_disabled("Log Local Runqueue");
 #endif
+
+if (USING_EARLIEST_START_FIRST) {
+	pretty_print_key_enabled("USING_EARLIEST_START_FIRST");
+} else {
+	pretty_print_key_disabled("USING_EARLIEST_START_FIRST");
+}
 }
 
 void
@@ -491,6 +515,7 @@ main(int argc, char **argv)
 	printf("Runtime Environment:\n");
 
 	runtime_set_resource_limits_to_max();
+	runtime_set_policy_and_prio();
 	runtime_allocate_available_cores();
 	runtime_configure();
 	runtime_initialize();
@@ -500,6 +525,11 @@ main(int argc, char **argv)
 	runtime_start_runtime_worker_threads();
 	runtime_get_processor_speed_MHz();
 	runtime_configure_worker_spinloop_pause();
+
+	runtime_quantum = runtime_quantum_us * runtime_processor_speed_MHz;
+	pretty_print_key_value("Quantum", "%u us = %lu cycles\n", runtime_quantum_us, runtime_quantum);
+
+	// traffic_control_initialize();
 	software_interrupt_arm_timer();
 
 #ifdef LOG_TENANT_LOADING
@@ -524,11 +554,16 @@ main(int argc, char **argv)
 			exit(-1);
 		}
 
+		if (runtime_max_deadline < tenant->max_relative_deadline) runtime_max_deadline = tenant->max_relative_deadline;
+
 		/* Start listening for requests */
 		rc = tenant_listen(tenant);
 		if (rc < 0) exit(-1);
 	}
-
+#ifdef TRAFFIC_CONTROL
+	traffic_control_initialize();
+	tenant_database_init_reservations();
+#endif
 	runtime_boot_timestamp = __getcycles();
 
 	for (int tenant_idx = 0; tenant_idx < tenant_config_vec_len; tenant_idx++) {

runtime/src/module.c

@@ -44,8 +44,9 @@ module_init(struct module *module, char *path)
 	rc = sledge_abi_symbols_init(&module->abi, path);
 	if (rc != 0) goto err;
 
-	module->pools = calloc((module->type == APP_MODULE ? runtime_worker_threads_count : 1),
-	                       sizeof(struct module_pool));
+	const int n = module->type == APP_MODULE ? runtime_worker_threads_count : 1;
+	// module->pools = calloc(n, sizeof(struct module_pool));
+	module->pools = calloc(1, sizeof(struct module_pool));
 
 	module->path = path;
 

runtime/src/runtime.c

@@ -10,10 +10,13 @@
 #include <sys/time.h>
 
 #include "admissions_control.h"
+#include "traffic_control.h"
 #include "arch/context.h"
 #include "debuglog.h"
 #include "global_request_scheduler_deque.h"
 #include "global_request_scheduler_minheap.h"
+#include "global_request_scheduler_mtds.h"
+#include "global_request_scheduler_mtdbf.h"
 #include "http_parser_settings.h"
 #include "listener_thread.h"
 #include "module.h"
@@ -32,6 +35,9 @@ int       *runtime_worker_threads_argument;
 /* The active deadline of the sandbox running on each worker thread */
 uint64_t *runtime_worker_threads_deadline;
 
+/* Tracks alive sandboxes */
+bool sandbox_refs[RUNTIME_MAX_ALIVE_SANDBOXES] = { false };
+
 /******************************************
  * Shared Process / Listener Thread Logic *
  *****************************************/
@@ -197,3 +203,35 @@ runtime_set_prio(unsigned int nice)
 
 	return;
 }
+
+void
+runtime_set_policy_and_prio()
+{
+	if (geteuid() != 0) {
+		printf("Won't set the priority for sledgert. Run with sudo again.\n");
+		return;
+	}
+
+	// int which = PRIO_PROCESS; // You can also use PRIO_PGRP or PRIO_USER
+    // int nice_value = -20;     // Adjust the nice value as needed
+
+    // if (setpriority(which, getpid(), nice_value) == -1) {
+    //     perror("setpriority");
+    // }
+
+	const int FIFO_PRIO = 1; //sched_get_priority_max(SCHED_FIFO);
+    struct sched_param sp;
+
+    // Set the scheduling policy to SCHED_FIFO
+    sp.sched_priority = FIFO_PRIO;
+
+	int rc = sched_setscheduler(getpid(), SCHED_FIFO, &sp);
+    if (rc != 0) {
+        panic("sched_setscheduler error. Return code: %d", rc);
+    }
+
+	if (sched_getparam(0, &sp) < 0) { perror("getparam: "); }
+	assert(sp.sched_priority == FIFO_PRIO);
+
+	printf("Set sched for sledgert to FIFO. Prio=%d\n", FIFO_PRIO);
+}

runtime/src/sandbox.c

@@ -2,6 +2,7 @@
 #include <string.h>
 #include <sys/mman.h>
 
+#include "sandbox_types.h"
 #include "current_sandbox.h"
 #include "debuglog.h"
 #include "panic.h"
@@ -38,6 +39,10 @@ sandbox_allocate_linear_memory(struct sandbox *sandbox)
 	sandbox->memory = module_allocate_linear_memory(sandbox->module);
 	if (unlikely(sandbox->memory == NULL)) return -1;
 
+	// sandbox->sizes[sandbox->sizesize] = sandbox->memory->abi.size;
+	// sandbox->sizesize++;
+	sandbox->memory->abi.id = sandbox->id;
+
 	return 0;
 }
 
@@ -77,9 +82,22 @@ sandbox_free_stack(struct sandbox *sandbox)
 {
 	assert(sandbox);
 
-	return module_free_stack(sandbox->module, sandbox->stack);
+	return module_free_stack(sandbox->module, sandbox->stack, sandbox->original_owner_worker_idx);
 }
 
+// /**
+//  * Free Linear Memory, leaving stack in place
+//  * @param sandbox
+//  */
+// static inline void
+// sandbox_free_linear_memory(struct sandbox *sandbox)
+// {
+// 	assert(sandbox != NULL);
+// 	assert(sandbox->memory != NULL);
+// 	module_free_linear_memory(sandbox->module, (struct wasm_memory *)sandbox->memory);
+// 	sandbox->memory = NULL;
+// }
+
 /**
  * Allocates HTTP buffers and performs our approximation of "WebAssembly instantiation"
  * @param sandbox
@@ -129,15 +147,15 @@ err_stack_allocation_failed:
 err_memory_allocation_failed:
 err_globals_allocation_failed:
 err_http_allocation_failed:
-	sandbox_set_as_error(sandbox, SANDBOX_ALLOCATED);
+	sandbox_set_as_error(sandbox, SANDBOX_INITIALIZED);
+	sandbox_free(sandbox);
 	perror(error_message);
 	rc = -1;
 	goto done;
 }
 
 void
-sandbox_init(struct sandbox *sandbox, struct module *module, struct http_session *session, struct route *route,
-             struct tenant *tenant, uint64_t admissions_estimate)
+sandbox_init(struct sandbox *sandbox, struct module *module, struct http_session *session, uint64_t admissions_estimate)
 {
 	/* Sets the ID to the value before the increment */
 	sandbox->id     = sandbox_total_postfix_increment();
@@ -150,12 +168,19 @@ sandbox_init(struct sandbox *sandbox, struct module *module, struct http_session
 	/* Allocate HTTP session structure */
 	assert(session);
 	sandbox->http   = session;
-	sandbox->tenant = tenant;
-	sandbox->route  = route;
+	sandbox->tenant = session->tenant;
+	sandbox->route  = session->route;
 
 	sandbox->absolute_deadline = sandbox->timestamp_of.allocation + sandbox->route->relative_deadline;
 	sandbox->payload_size      = session->http_request.body_length;
 
+	sandbox->exceeded_estimation = false;
+	sandbox->writeback_preemption_in_progress = false;
+	sandbox->writeback_overshoot_in_progress = false;
+	sandbox->response_code       = 0;
+	sandbox->owned_worker_idx    = -2;
+	sandbox->original_owner_worker_idx = -2;
+
 	/*
 	 * Admissions Control State
 	 * Assumption: an estimate of 0 should have been interpreted as a rejection
@@ -176,8 +201,7 @@ sandbox_init(struct sandbox *sandbox, struct module *module, struct http_session
  * @return the new sandbox request
  */
 struct sandbox *
-sandbox_alloc(struct module *module, struct http_session *session, struct route *route, struct tenant *tenant,
-              uint64_t admissions_estimate)
+sandbox_alloc(struct module *module, struct http_session *session, uint64_t admissions_estimate, uint64_t sandbox_alloc_timestamp)
 {
 	size_t alignment     = (size_t)PAGE_SIZE;
 	size_t size_to_alloc = (size_t)round_up_to_page(sizeof(struct sandbox));
@@ -190,13 +214,35 @@ sandbox_alloc(struct module *module, struct http_session *session, struct route
 	if (unlikely(sandbox == NULL)) return NULL;
 	memset(sandbox, 0, size_to_alloc);
 
+	sandbox->timestamp_of.allocation = sandbox_alloc_timestamp;
 	sandbox_set_as_allocated(sandbox);
-	sandbox_init(sandbox, module, session, route, tenant, admissions_estimate);
+	sandbox_init(sandbox, module, session, admissions_estimate);
 
+	sandbox_refs[sandbox->id % RUNTIME_MAX_ALIVE_SANDBOXES] = true;
 
 	return sandbox;
 }
 
+struct sandbox_metadata *sandbox_meta_alloc(struct sandbox *sandbox)
+{
+	struct sandbox_metadata *sandbox_meta = malloc(sizeof(struct sandbox_metadata));
+
+	sandbox_meta->sandbox_shadow       = sandbox;
+	sandbox_meta->tenant               = sandbox->tenant;
+	sandbox_meta->route                = sandbox->route;
+	sandbox_meta->id                   = sandbox->id;
+	sandbox_meta->state                = sandbox->state;
+	sandbox_meta->allocation_timestamp = sandbox->timestamp_of.allocation;
+	sandbox_meta->absolute_deadline    = sandbox->absolute_deadline;
+	sandbox_meta->remaining_exec       = sandbox->remaining_exec;
+	sandbox_meta->exceeded_estimation  = sandbox->exceeded_estimation;
+	sandbox_meta->owned_worker_idx     = worker_thread_idx;
+	sandbox_meta->terminated           = false;
+	sandbox_meta->pq_idx_in_tenant_queue = 0;
+
+	return sandbox_meta;
+}
+
 void
 sandbox_deinit(struct sandbox *sandbox)
 {
@@ -210,8 +256,9 @@ sandbox_deinit(struct sandbox *sandbox)
 	module_release(sandbox->module);
 
 	/* Linear Memory and Guard Page should already have been munmaped and set to NULL */
-	assert(sandbox->memory == NULL);
+	// assert(sandbox->memory == NULL);
 
+	if (likely(sandbox->memory != NULL)) sandbox_free_linear_memory(sandbox);
 	if (likely(sandbox->stack != NULL)) sandbox_free_stack(sandbox);
 	if (likely(sandbox->globals.buffer != NULL)) sandbox_free_globals(sandbox);
 	if (likely(sandbox->wasi_context != NULL)) wasi_context_destroy(sandbox->wasi_context);
@@ -227,7 +274,9 @@ sandbox_free(struct sandbox *sandbox)
 	assert(sandbox != NULL);
 	assert(sandbox != current_sandbox_get());
 	assert(sandbox->state == SANDBOX_ERROR || sandbox->state == SANDBOX_COMPLETE);
+	assert(!listener_thread_is_running() || sandbox->memory == NULL);
 
+	sandbox_refs[sandbox->id % RUNTIME_MAX_ALIVE_SANDBOXES] = false;
 	sandbox_deinit(sandbox);
 	free(sandbox);
 }

runtime/src/scheduler.c

@@ -1,3 +1,172 @@
 #include "scheduler.h"
 
 enum SCHEDULER scheduler = SCHEDULER_EDF;
+
+void
+sandbox_process_scheduler_updates(struct sandbox *sandbox)
+{
+	if (scheduler == SCHEDULER_MTDS && tenant_is_paid(sandbox->tenant)) {
+		atomic_fetch_sub(&sandbox->tenant->remaining_budget, sandbox->last_running_state_duration);
+		sandbox->last_running_state_duration = 0;
+		return;
+	}
+
+#ifdef TRAFFIC_CONTROL
+	assert(sandbox->sandbox_meta);
+	assert(sandbox == sandbox->sandbox_meta->sandbox_shadow);
+	assert(sandbox->id == sandbox->sandbox_meta->id);
+
+	struct comm_with_worker *cfw = &comm_from_workers[worker_thread_idx];
+	assert(cfw);
+
+	const uint64_t now = __getcycles();
+
+	struct message new_message = {
+		.sandbox                     = sandbox,
+		.sandbox_id                  = sandbox->id,
+		.sandbox_meta                = sandbox->sandbox_meta,
+		.state                       = sandbox->state,
+		.sender_worker_idx           = worker_thread_idx,
+		.exceeded_estimation         = sandbox->exceeded_estimation,
+		.total_running_duration      = 0,
+		.timestamp = now
+	};
+
+	if (sandbox->state == SANDBOX_RETURNED || sandbox->state == SANDBOX_ERROR) {
+		uint64_t adjustment = sandbox->last_running_state_duration;
+		if (sandbox->remaining_exec < adjustment) adjustment = sandbox->remaining_exec;
+		// const uint64_t adjustment = sandbox->remaining_exec;
+
+		if (USING_LOCAL_RUNQUEUE && adjustment > 0 && sandbox->response_code == 0) {
+			// dbf_try_update_demand(worker_dbf, sandbox->timestamp_of.dispatched,
+			//                       sandbox->route->relative_deadline, sandbox->absolute_deadline, sandbox->remaining_exec,
+			//                       DBF_DELETE_EXISTING_DEMAND, NULL, NULL);
+		}
+		
+		// sandbox->remaining_exec = 0;
+
+		new_message.message_type        = MESSAGE_CFW_DELETE_SANDBOX;
+		new_message.adjustment          = adjustment;
+		// new_message.remaining_exec = 0;
+		new_message.remaining_exec = sandbox->remaining_exec;
+		new_message.total_running_duration = sandbox->duration_of_state[SANDBOX_RUNNING_USER] + sandbox->duration_of_state[SANDBOX_RUNNING_SYS];
+
+		if (!ck_ring_enqueue_spsc_message(&cfw->worker_ring, cfw->worker_ring_buffer, &new_message)) {
+			panic("Ring The buffer was full and the enqueue operation has failed.!");
+		}
+
+		return;
+	} 
+
+	/* Unless the sandbox is in the terminal state (handled above), then the only state it can be is INTERRUPTED */
+	assert(sandbox->state == SANDBOX_INTERRUPTED);
+	assert(sandbox == current_sandbox_get());
+	assert(sandbox->response_code == 0);
+	assert(sandbox->remaining_exec > 0);
+	assert(!sandbox->exceeded_estimation || sandbox->remaining_exec == runtime_quantum);
+
+	if (sandbox->sandbox_meta->terminated) {
+        assert(sandbox->sandbox_meta->error_code > 0);
+		// dbf_try_update_demand(worker_dbf, sandbox->timestamp_of.dispatched,
+		// 	                      sandbox->route->relative_deadline, sandbox->absolute_deadline, sandbox->remaining_exec,
+		// 	                      DBF_DELETE_EXISTING_DEMAND, NULL, NULL);
+		sandbox->response_code = sandbox->sandbox_meta->error_code;
+		interrupted_sandbox_exit();
+		return;
+	}
+
+	if (sandbox->absolute_deadline < now + (!sandbox->exceeded_estimation ? sandbox->remaining_exec : 0)) {
+		// dbf_try_update_demand(worker_dbf, sandbox->timestamp_of.dispatched,
+		// 	                      sandbox->route->relative_deadline, sandbox->absolute_deadline, sandbox->remaining_exec,
+		// 	                      DBF_DELETE_EXISTING_DEMAND, NULL, NULL);
+		sandbox->response_code = 4081;
+		interrupted_sandbox_exit();
+		return;
+	}
+	
+	dbf_update_mode_t dbf_reduce_mode = DBF_REDUCE_EXISTING_DEMAND;
+	uint64_t adjustment = sandbox->last_running_state_duration;
+	if (sandbox->remaining_exec < sandbox->last_running_state_duration || sandbox->exceeded_estimation) {
+		/* To avoid less than quantum updates manually set the adjustment to quantum */
+		adjustment = sandbox->remaining_exec;
+		dbf_reduce_mode = DBF_DELETE_EXISTING_DEMAND;
+	}
+
+	sandbox->last_running_state_duration = 0;
+	sandbox->remaining_exec   -= adjustment;
+	
+	new_message.adjustment          = adjustment;
+	new_message.message_type        = MESSAGE_CFW_REDUCE_DEMAND;
+	new_message.remaining_exec = sandbox->remaining_exec;
+
+	if (USING_LOCAL_RUNQUEUE /* && !sandbox->exceeded_estimation */) {
+		// dbf_try_update_demand(worker_dbf, sandbox->timestamp_of.dispatched,
+		// 						sandbox->route->relative_deadline, sandbox->absolute_deadline, adjustment,
+		// 						dbf_reduce_mode, NULL, NULL);
+	}
+
+	if (!ck_ring_enqueue_spsc_message(&cfw->worker_ring, cfw->worker_ring_buffer, &new_message)) {
+		panic("Ring The buffer was full and the enqueue operation has failed.!")
+	}
+
+	if (sandbox->remaining_exec == 0) {
+		/* OVERSHOOT case! */
+		// printf("Went over estimation - sandbox_id=%lu of %s!\n", sandbox->id, sandbox->tenant->name);
+		if (sandbox->exceeded_estimation == false) sandbox->tenant->num_of_overshooted_sandboxes++;
+		sandbox->exceeded_estimation = true;
+		sandbox->num_of_overshoots++;
+		if (sandbox->num_of_overshoots > sandbox->tenant->max_overshoot_of_same_sandbox) {
+			sandbox->tenant->max_overshoot_of_same_sandbox = sandbox->num_of_overshoots;
+		}
+		
+		const uint64_t extra_demand = runtime_quantum;
+
+		if (USING_LOCAL_RUNQUEUE && USING_TRY_LOCAL_EXTRA 
+			/*&& dbf_try_update_demand(worker_dbf, now, sandbox->route->relative_deadline,
+		                          sandbox->absolute_deadline, extra_demand, DBF_CHECK_AND_ADD_DEMAND, &new_message, NULL)*/
+								|| (!USING_LOCAL_RUNQUEUE && USING_TRY_LOCAL_EXTRA)) {
+			/* Worker DBF has supply left */
+			// printf("Worker %d granted extra for sandbox %lu!\n", worker_thread_idx, sandbox->id);
+
+			sandbox->remaining_exec = extra_demand;
+
+			new_message.adjustment = extra_demand;
+			new_message.exceeded_estimation = true;
+			new_message.message_type = MESSAGE_CFW_EXTRA_DEMAND_REQUEST;
+			new_message.remaining_exec = extra_demand;
+
+			if (!ck_ring_enqueue_spsc_message(&cfw->worker_ring, cfw->worker_ring_buffer, &new_message)) {
+				panic("Ring The buffer was full and the enqueue operation has failed.!")
+			}
+
+			return;
+		} else if (USING_WRITEBACK_FOR_OVERSHOOT) { 
+			/* Write back */
+			// printf("No supply left in worker #%d. So, writeback sandbox=%lu of %s\n", worker_thread_idx, sandbox->id, sandbox->tenant->name);
+
+			sandbox->remaining_exec = 0;
+			sandbox->writeback_overshoot_in_progress= true;
+			local_runqueue_delete(sandbox); // TODO: This needs to go in preemp_sandbox state change!
+			return;
+		} else { 
+			/* Kill work */
+			// printf("No supply left in worker #%d. So, kill sandbox=%lu of %s\n", worker_thread_idx, sandbox->id, sandbox->tenant->name);
+
+            assert(sandbox->response_code == 0);
+			sandbox->response_code = 4093;
+			interrupted_sandbox_exit();
+			return;
+		}
+	}
+
+#else
+
+	if (sandbox->remaining_exec > sandbox->last_running_state_duration) {
+		sandbox->remaining_exec -= sandbox->last_running_state_duration;
+	} else {
+		sandbox->remaining_exec = 0;
+	}
+	sandbox->last_running_state_duration = 0;
+
+#endif
+}

runtime/src/sledge_abi.c

@@ -43,14 +43,25 @@ check_bounds(uint32_t offset, uint32_t bounds_check)
 EXPORT int32_t
 sledge_abi__wasm_memory_expand(struct sledge_abi__wasm_memory *wasm_memory, uint32_t page_count)
 {
+	uint64_t oldsize = wasm_memory->size;
 	struct sandbox *sandbox = current_sandbox_get();
 
+assert(sandbox->sandbox_meta);
+if (!(sandbox->id == wasm_memory->id)) fprintf(stderr, "sand_id: %lu, wasm_id: %lu, terminated: %u, orig_idx: %d, owner_wrk; %d, wrk: %d\n", sandbox->id, wasm_memory->id, sandbox->sandbox_meta->terminated, sandbox->original_owner_worker_idx, sandbox->owned_worker_idx, worker_thread_idx);
+assert(sandbox->id == wasm_memory->id);
+
+// if (!(sandbox->memory->abi.size == wasm_memory->size)) printf("sand_memsize: %lu, wasm_size: %lu, wasm_oldsize: %lu, targetsize: %lu\n", sandbox->memory->abi.size, wasm_memory->size, oldsize, page_count * WASM_PAGE_SIZE);
+// assert(sandbox->memory->abi.size == wasm_memory->size);
+
+if (sandbox->state != SANDBOX_RUNNING_USER) panic ("state!!! %u\n", sandbox->state);
+
 	sandbox_syscall(sandbox);
 	int32_t old_page_count = wasm_memory->size / WASM_PAGE_SIZE;
 	int     rc             = wasm_memory_expand((struct wasm_memory *)wasm_memory, page_count * WASM_PAGE_SIZE);
 
 	if (unlikely(rc == -1)) {
 		old_page_count = -1;
+		assert(0);
 		goto DONE;
 	}
 
@@ -58,6 +69,13 @@ sledge_abi__wasm_memory_expand(struct sledge_abi__wasm_memory *wasm_memory, uint
 	 * the original struct as well  */
 	current_sandbox_memory_writeback();
 
+	
+	
+	// if (!(sandbox->memory->abi.size == wasm_memory->size)) printf("sand_memsize: %lu, wasm_size: %lu, wasm_oldsize: %lu, targetsize: %lu\n", sandbox->memory->abi.size, wasm_memory->size, oldsize, page_count * WASM_PAGE_SIZE);
+	// assert(sandbox->memory->abi.size == wasm_memory->size);
+	// sandbox->sizes[sandbox->sizesize] = wasm_memory->size + page_count * WASM_PAGE_SIZE;
+	// sandbox->sizesize++;
+
 #ifdef LOG_SANDBOX_MEMORY_PROFILE
 	// Cache the runtime of the first N page allocations
 	for (int i = 0; i < page_count; i++) {

runtime/src/software_interrupt.c

@@ -179,6 +179,7 @@ software_interrupt_handle_signals(int signal_type, siginfo_t *signal_info, void
 			atomic_fetch_sub(&handler_depth, 1);
 			current_sandbox_trap(WASM_TRAP_OUT_OF_BOUNDS_LINEAR_MEMORY);
 		} else {
+			assert(0);
 			panic("Runtime SIGSEGV\n");
 		}
 

runtime/src/tenant_database.c

@@ -3,6 +3,8 @@
 #include "panic.h"
 #include "runtime.h"
 #include "tenant.h"
+#include "panic.h"
+#include "tenant_functions.h"
 
 /*******************
  * Tenant Database *
@@ -35,7 +37,6 @@ err_no_space:
 	goto done;
 }
 
-
 /**
  * Given a name, find the associated tenant
  * @param name
@@ -102,3 +103,138 @@ tenant_database_foreach(void (*cb)(struct tenant *, void *), void *arg)
 		cb(tenant_database[i], arg);
 	}
 }
+
+#ifdef TRAFFIC_CONTROL
+void
+tenant_database_init_reservations()
+{
+	printf("Runtime Max Deadline: %lu\n", runtime_max_deadline);
+
+	for (size_t i = 0; i < tenant_database_count; i++) {
+		assert(tenant_database[i]);
+		struct tenant *t = tenant_database[i];
+		if (t->tcp_server.port == 55555) continue;
+
+		t->trs.max_budget_guaranteed = REPLENISHMENT_PERIOD * runtime_worker_threads_count * t->reservation_percentile / 100;
+		t->trs.budget_guaranteed = t->trs.max_budget_guaranteed;
+		t->trs.budget_best = UINT64_MAX;
+
+		printf("Tenant %s, max_deadline: %lu, max_budget: %ld\n", t->name, t->max_relative_deadline, t->trs.max_budget_guaranteed);
+	}
+}
+
+/**
+ * @brief Iterate through the tenant databse and print DBF info per tenant
+ */
+void
+tenant_database_print_reservations()
+{
+	for (size_t i = 0; i < tenant_database_count; i++) {
+		assert(tenant_database[i]);
+		struct tenant *t = tenant_database[i];
+		if (t->tcp_server.port == 55555) continue;
+
+		printf("\nTENANT: %s INFO:\n", t->name);
+		printf("Global_meta_size: %d, Local_meta_size: %d\n", priority_queue_length_nolock(t->global_sandbox_metas), priority_queue_length_nolock(t->local_sandbox_metas));
+		printf("Number of total overshoots: %u, MAX of overshoots from the same sandbox: %u \n", t->num_of_overshooted_sandboxes, t->max_overshoot_of_same_sandbox);
+		// printf("Best Effort demands: %lu\n", t->trs.best_effort_cycles);
+		// dbf_print(t->tenant_dbf, __getcycles());
+		
+		tenant_print_jobs(t);
+	}
+}
+
+/**
+ * Checks is an opaque pointer is a tenant by comparing against
+ */
+struct tenant *
+tenant_database_find_tenant_most_oversupply(struct tenant *tenant_to_exclude, uint64_t time_of_oversupply, bool weak_shed, struct sandbox_metadata **sandbox_meta_to_remove)
+{
+	assert(sandbox_meta_to_remove != NULL);
+	struct tenant *tenant_to_punish    = NULL;
+	uint64_t       min_tenant_BE_budget = UINT64_MAX;
+
+	for (size_t i = 0; i < tenant_database_count; i++) {
+		assert(tenant_database[i]);
+		// if (tenant_database[i] == tenant_to_exclude) continue;
+		struct tenant *tenant = tenant_database[i];
+		assert(tenant);
+		if (tenant->tcp_server.port == 55555) continue;
+		if (priority_queue_length_nolock(tenant->global_sandbox_metas) + priority_queue_length_nolock(tenant->local_sandbox_metas) == 0) continue;
+
+		struct sandbox_metadata *sandbox_meta = NULL;
+		int rc = priority_queue_top_nolock(tenant->global_sandbox_metas, (void **)&sandbox_meta);
+
+		if (!sandbox_meta || sandbox_meta->absolute_deadline > time_of_oversupply) {
+			sandbox_meta = NULL;
+			rc = priority_queue_top_nolock(tenant->local_sandbox_metas, (void **)&sandbox_meta);
+		}
+		if (!sandbox_meta || sandbox_meta->absolute_deadline > time_of_oversupply) continue;
+		assert(rc == 0);
+		assert(sandbox_meta);
+
+		if (tenant->trs.budget_best < min_tenant_BE_budget) {
+			min_tenant_BE_budget = tenant->trs.budget_best;
+			tenant_to_punish     = tenant;
+			*sandbox_meta_to_remove = sandbox_meta;
+		}
+	}
+
+/*
+	if(weak_shed && tenant_to_punish == NULL) printf("Weak mode: No tenant to punish for %s\n", tenant_to_exclude->name);
+	if(!weak_shed && tenant_to_punish == NULL) printf("Strong mode: No tenant to punish for %s\n", tenant_to_exclude->name);
+
+	if (weak_shed && tenant_to_punish) {
+		assert(tenant_to_punish->reservation_percentile==20);
+		printf("MODE weak=%u - pending tenant %s \n", weak_shed, tenant_to_exclude->name);
+		printf("Start (ms): %lu\n", start_time);
+		printf("AbsDL (ms): %lu\n", absolute_deadline);
+		printf("ToS   (ms): %lu\n", time_of_oversupply);
+		printf("RelativeDL (ms): %lu\n", absolute_deadline/runtime_quantum - start_time/runtime_quantum);
+		
+		min_tenant_BE_budget = UINT64_MAX;
+		for (size_t i = 0; i < tenant_database_count; i++) {
+			assert(tenant_database[i]);
+			struct tenant *tenant = tenant_database[i];
+			assert(tenant);
+			if (tenant->tcp_server.port == 55555) continue;
+
+			uint64_t t_budget_BE = tenant->trs.budget_BE;
+			if (t_budget_BE < min_tenant_BE_budget) {
+				min_tenant_BE_budget = t_budget_BE;
+				tenant_to_punish     = tenant;
+			}
+
+			printf("Tenant: %s, rp=%u, min_tenant_BE_budget=%lu\n", tenant->name,
+			       tenant->reservation_percentile, min_tenant_BE_budget);
+			tenant_print_jobs(tenant);
+			printf("Tenant Globl SIZE: %d\n", priority_queue_length_nolock(tenant->global_sandbox_metas));
+			printf("Tenant Local SIZE: %d\n\n", priority_queue_length_nolock(tenant->local_sandbox_metas));
+		}
+
+		const int N_VIRT_WORKERS_DBF = USING_AGGREGATED_GLOBAL_DBF ? 1 : runtime_worker_threads_count;
+		for (int i = 0; i < N_VIRT_WORKERS_DBF; i++) {
+			printf("GL Worker #%d\n", i);
+			dbf_print(global_virt_worker_dbfs[i], start_time);
+		}
+		assert(0);
+	}*/
+	assert((*sandbox_meta_to_remove)==NULL || (*sandbox_meta_to_remove)->tenant == tenant_to_punish);
+	return tenant_to_punish;
+}
+
+void
+tenant_database_replenish_all()
+{
+	const uint64_t now = __getcycles();
+
+	for (size_t i = 0; i < tenant_database_count; i++) {
+		assert(tenant_database[i]);
+		struct tenant *t = tenant_database[i];
+		if (t->tcp_server.port == 55555) continue;
+
+		tenant_replenish(t, now);
+	}
+}
+
+#endif

runtime/src/traffic_control.c

@@ -0,0 +1,254 @@
+#include <unistd.h>
+
+#include "traffic_control.h"
+#include "debuglog.h"
+#include "global_request_scheduler_mtdbf.h"
+#include "tenant_functions.h"
+#include "sandbox_set_as_error.h"
+#include "dbf.h"
+
+#ifdef TRAFFIC_CONTROL
+// void *global_dbf;
+void **global_virt_worker_dbfs;
+void *global_worker_dbf; // temp ///////////
+
+extern struct priority_queue *global_request_scheduler_mtdbf;//, *global_default;
+extern lock_t global_lock;
+
+void
+traffic_control_initialize()
+{
+	assert(runtime_max_deadline > 0);
+
+	const int N_VIRT_WORKERS_DBF = USING_AGGREGATED_GLOBAL_DBF ? 1 : runtime_worker_threads_count;
+	global_virt_worker_dbfs = malloc(N_VIRT_WORKERS_DBF * sizeof(void*));
+	for (int i = 0; i < N_VIRT_WORKERS_DBF; i++) {
+		global_virt_worker_dbfs[i] = dbf_initialize(runtime_worker_threads_count/N_VIRT_WORKERS_DBF, 100, -1, NULL);
+	}
+
+}
+
+void
+traffic_control_log_decision(const int num, const bool admitted)
+{
+#ifdef LOG_TRAFFIC_CONTROL
+	debuglog("Admission case #: %d, Admitted? %s\n", num, admitted ? "yes" : "no");
+#endif /* LOG_TRAFFIC_CONTROL */
+}
+int ind = 0;
+int
+global_virt_worker_dbfs_try_update_demand(uint64_t start_time, uint64_t adjustment, uint64_t *time_oversupply_p, struct sandbox_metadata *sm)
+{
+	bool global_can_admit = false;
+	uint64_t time_oversupply = 0;
+	const uint64_t absolute_deadline = sm->absolute_deadline;
+	const int N_VIRT_WORKERS_DBF = USING_AGGREGATED_GLOBAL_DBF ? 1 : runtime_worker_threads_count;
+
+	/* Hack the start time to make sure demand less than the quantum is also served */
+	if((absolute_deadline - start_time) * N_VIRT_WORKERS_DBF < runtime_quantum) start_time = absolute_deadline - runtime_quantum;
+
+	for (int i = ind; i < (N_VIRT_WORKERS_DBF) + ind; i++) {
+		assert(global_virt_worker_dbfs);
+		void *global_dbf = global_virt_worker_dbfs[i%N_VIRT_WORKERS_DBF];
+		global_can_admit = dbf_list_try_add_new_demand(global_dbf, start_time, absolute_deadline, adjustment, sm);
+		if (global_can_admit) {
+			ind = (i+1)%N_VIRT_WORKERS_DBF;
+			return i%N_VIRT_WORKERS_DBF;
+		}
+		
+		if (time_oversupply < dbf_get_time_of_oversupply(global_dbf)) time_oversupply =  dbf_get_time_of_oversupply(global_dbf);
+	}
+	
+	*time_oversupply_p = time_oversupply;
+	return -1;
+}
+
+uint64_t
+traffic_control_decide(struct sandbox_metadata *sandbox_meta, const uint64_t start_time, const uint64_t estimated_execution, int *ret_code, int *worker_id_virtual)
+{
+	/* Nominal non-zero value in case traffic control is disabled */
+	uint64_t work_admitted = estimated_execution;
+
+	int rc = 0;
+	int worker_id_v = -1;
+
+	assert(sandbox_meta);
+	struct tenant *tenant = sandbox_meta->tenant;
+	const uint64_t absolute_deadline = sandbox_meta->absolute_deadline;
+
+	uint64_t time_global_oversupply = 0;
+	worker_id_v = global_virt_worker_dbfs_try_update_demand(start_time, estimated_execution, &time_global_oversupply, sandbox_meta);
+	bool global_can_admit = worker_id_v >= 0;
+
+	// bool tenant_can_admit = tenant_try_add_job(tenant, start_time, estimated_execution, TRS_CHECK_GUARANTEED, sandbox_meta);
+	bool tenant_can_admit = tenant_can_admit_guaranteed(tenant, start_time, estimated_execution);
+
+	if (tenant_can_admit && global_can_admit) {
+		/* Case #1: Both the tenant and overall system is under utlized. So, just admit. */
+		tenant_can_admit = tenant_try_add_job_as_guaranteed(tenant, start_time, estimated_execution, sandbox_meta);
+		assert(tenant_can_admit);
+		traffic_control_log_decision(1, true);
+		rc = 1;
+	} else if (!tenant_can_admit && global_can_admit) {
+		/* Case #2: Tenant is over utilized, but system is under utilized. So, admit for work-conservation. */
+		if (USING_WORK_CONSERVATION == false) {
+			traffic_control_log_decision(2, false);
+			dbf_try_update_demand(global_virt_worker_dbfs[worker_id_v], start_time,
+												0, absolute_deadline, estimated_execution,
+												DBF_DELETE_EXISTING_DEMAND, NULL, sandbox_meta);
+			goto any_work_not_admitted;
+		}
+
+		traffic_control_log_decision(2, true);
+		rc = 2;
+	} else if (tenant_can_admit && !global_can_admit) {
+		/* Case #3: Tenant is under utilized, but  system is over utilized. So, shed work and then admit. */
+		assert(time_global_oversupply >= absolute_deadline);
+		
+		int worker_id_virt_just_shed;
+		while (!global_can_admit) {
+			assert(worker_id_v < 0);
+
+			worker_id_virt_just_shed = -1;
+			uint64_t cleared_demand = traffic_control_shed_work(tenant, time_global_oversupply, &worker_id_virt_just_shed, false);
+			if (cleared_demand == 0) {
+				/* No "bad" tenant requests left in the global queue, so we have deny the guaranteed tenant job. */
+				traffic_control_log_decision(3, false);
+				goto guaranteed_work_not_admitted;
+			}
+
+			assert(worker_id_virt_just_shed >= 0);
+			void *global_dbf = global_virt_worker_dbfs[worker_id_virt_just_shed];
+			global_can_admit = dbf_list_try_add_new_demand(global_dbf, start_time, absolute_deadline, estimated_execution, sandbox_meta);
+			time_global_oversupply = dbf_get_time_of_oversupply(global_dbf);
+		}
+
+		worker_id_v = worker_id_virt_just_shed;
+		tenant_can_admit = tenant_try_add_job_as_guaranteed(tenant, start_time, estimated_execution, sandbox_meta);
+		assert(tenant_can_admit);
+		traffic_control_log_decision(3, true);
+		rc = 1;
+	} else if (!tenant_can_admit && !global_can_admit) {
+		/* Case #4: Do NOT admit. */
+
+		// printf("Case #4: Do NOT admit.\n");
+		// traffic_control_log_decision(4, false);
+		// goto any_work_not_admitted;
+
+		// assert(time_global_oversupply >= absolute_deadline);
+
+		int worker_id_virt_just_shed;
+		while (!global_can_admit) {
+			assert(worker_id_v < 0);
+
+			worker_id_virt_just_shed = -1;
+
+			uint64_t cleared_demand = traffic_control_shed_work(tenant, time_global_oversupply, &worker_id_virt_just_shed, true);
+			if (cleared_demand == 0) {
+				/* No "bad" tenant requests left in the global queue, so we have deny this new job. */
+				traffic_control_log_decision(4, false);
+				goto any_work_not_admitted;
+			}
+
+			assert(worker_id_virt_just_shed >= 0);
+			void *global_dbf = global_virt_worker_dbfs[worker_id_virt_just_shed];
+			global_can_admit = dbf_list_try_add_new_demand(global_dbf, start_time, absolute_deadline, estimated_execution, sandbox_meta);
+			time_global_oversupply = dbf_get_time_of_oversupply(global_dbf);
+		}
+// printf("Case #4: Do admit %s.\n", tenant->name);
+		assert (global_can_admit);
+		worker_id_v = worker_id_virt_just_shed;
+		rc = 2;
+	}
+
+done:
+	*ret_code = rc;
+	*worker_id_virtual = worker_id_v;
+	return work_admitted;
+any_work_not_admitted:
+	work_admitted = 0;
+	rc = sandbox_meta->exceeded_estimation ? 4295 : 4290;
+	goto done;
+guaranteed_work_not_admitted:
+	work_admitted = 0;
+	rc = sandbox_meta->exceeded_estimation ? 4296: 4291;
+	goto done;
+}
+
+
+uint64_t traffic_control_shed_work(struct tenant *tenant_to_exclude, uint64_t time_of_oversupply, int *worker_id_virt_just_shed, bool weak_shed)
+{
+	uint64_t cleared_demand = 0;
+	*worker_id_virt_just_shed = -1;
+	struct sandbox_metadata *sandbox_meta = NULL;
+
+	struct tenant *tenant_to_punish = tenant_database_find_tenant_most_oversupply(tenant_to_exclude, time_of_oversupply, weak_shed, &sandbox_meta);
+	if (tenant_to_punish == NULL) {
+		// printf("null\n");
+		assert (sandbox_meta == NULL);
+		goto done;
+	}
+	if (tenant_to_punish == tenant_to_exclude) {
+		// printf("itself\n");
+		// TODO: Should be able to kill from itself???
+		goto done;
+	}
+
+	assert(sandbox_meta);
+	assert(sandbox_meta->tenant == tenant_to_punish);
+	assert(sandbox_meta->absolute_deadline <= time_of_oversupply);
+	assert(sandbox_meta->terminated == false);
+	assert(sandbox_meta->error_code == 0);
+
+	if (sandbox_meta->state == SANDBOX_INITIALIZED) {
+		assert(sandbox_meta->tenant_queue == tenant_to_punish->global_sandbox_metas);
+		sandbox_meta->error_code = 4090;
+		assert(sandbox_meta->owned_worker_idx == -2);
+	} else {
+		assert(sandbox_meta->tenant_queue == tenant_to_punish->local_sandbox_metas);
+		sandbox_meta->error_code = 4091;
+
+		struct message new_message = { 0 };
+		if (sandbox_meta->owned_worker_idx >= 0 && sandbox_refs[sandbox_meta->id % RUNTIME_MAX_ALIVE_SANDBOXES]) {
+			assert(comm_to_workers);
+			struct comm_with_worker *ctw = &comm_to_workers[sandbox_meta->owned_worker_idx];
+			assert(ctw);
+			assert(ctw->worker_idx == sandbox_meta->owned_worker_idx);
+			assert(ck_ring_size(&ctw->worker_ring) < LISTENER_THREAD_RING_SIZE);
+
+			new_message.sandbox_meta = sandbox_meta;
+			new_message.sandbox      = sandbox_meta->sandbox_shadow;
+			new_message.sandbox_id   = sandbox_meta->id;
+			new_message.message_type = MESSAGE_CTW_SHED_CURRENT_JOB;
+
+			if (!ck_ring_enqueue_spsc_message(&ctw->worker_ring, ctw->worker_ring_buffer, &new_message)) {
+				panic("Ring buffer was full and the enqueue failed!")
+			}
+			pthread_kill(runtime_worker_threads[sandbox_meta->owned_worker_idx], SIGALRM);
+		}
+	}
+
+	struct sandbox_metadata *sm_to_remove = NULL;
+	int rc = priority_queue_dequeue_nolock(sandbox_meta->tenant_queue, (void **)&sm_to_remove);
+	assert(rc == 0);
+	assert(sandbox_meta == sm_to_remove);
+
+	assert(sandbox_meta->trs_job_node == NULL);
+	assert(sandbox_meta->remaining_exec > 0);
+	assert(sandbox_meta->global_queue_type == 2);
+	assert(sandbox_meta->worker_id_virt>=0);
+
+	void *global_dbf = global_virt_worker_dbfs[sandbox_meta->worker_id_virt];
+	dbf_list_reduce_demand(sandbox_meta, sandbox_meta->remaining_exec + sandbox_meta->extra_slack, true);
+	sandbox_meta->demand_node = NULL;
+
+	cleared_demand = sandbox_meta->remaining_exec;
+	// sandbox_meta->remaining_exec = 0;
+	// sandbox_meta->extra_slack = 0;
+	*worker_id_virt_just_shed = sandbox_meta->worker_id_virt;
+	sandbox_meta->terminated = true;
+done:
+	return cleared_demand;
+}
+
+#endif /* TRAFFIC_CONTROL */

runtime/src/worker_thread.c

@@ -17,6 +17,8 @@
 #include "scheduler.h"
 #include "tenant_functions.h"
 #include "worker_thread.h"
+#include "priority_queue.h"
+#include "dbf.h"
 
 /***************************
  * Worker Thread State     *
@@ -30,6 +32,12 @@ thread_local int worker_thread_idx;
 
 /* Used to track tenants' timeouts */
 thread_local struct priority_queue *worker_thread_timeout_queue;
+
+/* Used to track worker's dbf */
+thread_local void *worker_dbf;
+thread_local struct sandbox_metadata *sandbox_meta;
+
+
 /***********************
  * Worker Thread Logic *
  **********************/
@@ -60,6 +68,22 @@ worker_thread_main(void *argument)
 	if (scheduler == SCHEDULER_MTDS) {
 		worker_thread_timeout_queue = priority_queue_initialize(RUNTIME_MAX_TENANT_COUNT, false,
 		                                                        tenant_timeout_get_priority);
+	} else if (scheduler == SCHEDULER_MTDBF) {
+#ifdef TRAFFIC_CONTROL
+		/* Initialize worker's dbf data structure */
+		/* To make sure global dbf reads out the max deadline in the system */
+		sleep(1);
+		
+		// worker_dbf = dbf_initialize(1, 100, worker_thread_idx, NULL);
+		// worker_dbf = dbf_grow(worker_dbf, dbf_get_max_relative_dl(global_dbf));
+		// worker_dbf = dbf_grow(worker_dbf, runtime_max_deadline);
+		// printf("WORKER ");
+		// dbf_print(worker_dbf);
+		sandbox_meta         = malloc(sizeof(struct sandbox_metadata));
+		sandbox_meta->tenant = NULL;
+
+		// if (worker_thread_idx == 0) global_worker_dbf = worker_dbf; // TODO: temp for debugging
+#endif
 	}
 
 	software_interrupt_unmask_signal(SIGFPE);

tests/bash_libraries/generate_spec_json.sh

@@ -87,7 +87,7 @@ generate_spec_json() {
 		done
 	done
 
-	if [ "$CLIENT_TERMINATE_SERVER" == true ]; then jq_admin_spec; fi
+	if [ "$ADMIN_ACCESS" == true ]; then jq_admin_spec; fi
 
 	# Merges all of the multiple specs for a single module
 	jq -s '. | sort_by(.name)' ./result_*.json > "./spec.json"

tests/common/mtdbf_preemption.env.bak

@@ -0,0 +1,4 @@
+SLEDGE_SCHEDULER=MTDBF
+SLEDGE_DISABLE_PREEMPTION=false
+SLEDGE_SANDBOX_PERF_LOG=perf.log
+SLEDGE_SPINLOOP_PAUSE_ENABLED=false

tests/mt-emil/.gitignore

@@ -0,0 +1 @@
+out.png

tests/mt-emil/Makefile

@@ -0,0 +1,102 @@
+SLEDGE_BINARY_DIR=../../runtime/bin
+HOST?=localhost # pass arguments to change this: make client-lpd HOST=10.10.1.4
+# HOST=arena0.andrew.cmu.edu
+# HOST=c220g2-011017.wisc.cloudlab.us
+PORT0=10000
+PORT1=15000
+PORT2=20000
+PORT3=25000
+PORT4=30000
+PORT5=35000
+PORT6=40000
+HEY_OPTS=-disable-compression -disable-keepalive -disable-redirects
+
+default: run
+
+clean: 
+	rm -rf res/*
+
+run:
+	SLEDGE_SIGALRM_HANDLER=TRIAGED SLEDGE_SCHEDULER=MTDBF SLEDGE_SPINLOOP_PAUSE_ENABLED=true SLEDGE_HTTP_SESSION_PERF_LOG=http_perf.log SLEDGE_SANDBOX_PERF_LOG=perf.log LD_LIBRARY_PATH=${SLEDGE_BINARY_DIR} ${SLEDGE_BINARY_DIR}/sledgert spec.json
+
+debug:
+	SLEDGE_SCHEDULER=MTDBF SLEDGE_SPINLOOP_PAUSE_ENABLED=false SLEDGE_NWORKERS=18 LD_LIBRARY_PATH=${SLEDGE_BINARY_DIR} gdb ${SLEDGE_BINARY_DIR}/sledgert \
+		--eval-command="handle SIGUSR1 noprint nostop" \
+		--eval-command="handle SIGPIPE noprint nostop" \
+		--eval-command="set pagination off" \
+		--eval-command="set print pretty" \
+		--eval-command="run spec.json"
+
+valgrind:
+	SLEDGE_DISABLE_PREEMPTION=true SLEDGE_NWORKERS=1 LD_LIBRARY_PATH=${SLEDGE_BINARY_DIR} valgrind --leak-check=full --max-stackframe=11150456 --run-libc-freeres=no --run-cxx-freeres=no ${SLEDGE_BINARY_DIR}/sledgert spec.json
+
+
+client-cnn:
+	curl -H 'Expect: ' -H "Content-Type: image/jpeg" --data-binary "@input-cnn/faces01.jpg" "${HOST}:${PORT0}/cnn"
+
+client-cifar10:
+	curl  -H 'Expect: ' -H "Content-Type: image/bmp" --data-binary "@input-cifar10/airplane1.bmp" "${HOST}:${PORT1}/cifar10"
+
+client-gocr:
+	curl  -H 'Expect: ' -H "Content-Type: application/octet-stream" --data-binary "@input-gocr/5x8.pnm" "${HOST}:${PORT2}/gocr"
+
+client-lpd:
+#	curl -H 'Expect: ' -H "Content-Type: image/png" --data-binary "@input-lpd-png/Cars0.png" "${HOST}:${PORT3}/lpd"
+	curl -H 'Expect: ' -H "Content-Type: image/jpeg" --data-binary "@input-lpd-jpg/Cars0.jpg" "${HOST}:${PORT3}/lpd"
+
+client-resize:
+	curl -H 'Expect: ' -H "Content-Type: image/jpeg" --data-binary "@input-resize/picsum_512x512_01.jpg" "${HOST}:${PORT4}/resize" --output "out-resize.jpg"
+
+client-ekf:
+	curl -H 'Expect: ' -H "Content-Type: application/octet-stream" --data-binary "@input-ekf/iter00.dat" "${HOST}:${PORT5}/ekf" --output "out-ekf-iter00.dat"
+
+client-fib-curl:
+	curl -i -H 'Expect: ' -H "Content-Type: text/plain" "${HOST}:${PORT6}/fib?30"
+
+########################################## Choose a random file to send with curl: ##########################################
+client-cnn-random:
+	@dir="input-cnn"; random_file="$$(ls $$dir | shuf -n 1)"; echo "Random file: $$random_file"; \
+	curl -s -H 'Expect: ' -H "Content-Type: image/jpeg" --data-binary "@$$dir/$$random_file" "${HOST}:${PORT0}/cnn"
+
+client-cifar10-random:
+	@dir="input-cifar10"; random_file="$$(ls $$dir | shuf -n 1)"; echo "Random file: $$random_file"; \
+	curl -s -H 'Expect: ' -H "Content-Type: image/bmp" --data-binary "@$$dir/$$random_file" "${HOST}:${PORT1}/cifar10"
+
+client-gocr-random:
+	@dir="input-gocr"; random_file="$$(ls $$dir | shuf -n 1)"; echo "Random file: $$random_file"; \
+	curl -s -H 'Expect: ' -H "Content-Type: application/octet-stream" --data-binary "@$$dir/$$random_file" "${HOST}:${PORT2}/gocr"
+
+client-lpd-random:
+#	@dir="input-lpd-png"; random_file="$$(ls $$dir | shuf -n 1)"; echo "Random file: $$random_file"; \
+#	curl -s -H 'Expect: ' -H "Content-Type: image/png" --data-binary "@$$dir/$$random_file" "${HOST}:${PORT3}/lpd"
+	@dir="input-lpd-jpg"; random_file="$$(ls $$dir | shuf -n 1)"; echo "Random file: $$random_file"; \
+	curl -s -H 'Expect: ' -H "Content-Type: image/jpeg" --data-binary "@$$dir/$$random_file" "${HOST}:${PORT3}/lpd"
+
+client-resize-random:
+	@dir="input-resize"; random_file="$$(ls $$dir | shuf -n 1)"; echo "Random file: $$random_file"; \
+	curl -s -H 'Expect: ' -H "Content-Type: image/jpeg" --data-binary "@$$dir/$$random_file" "${HOST}:${PORT4}/resize" --output "out-resize-$$random_file"
+
+client-ekf-random:
+	@dir="input-ekf"; random_file="$$(ls $$dir | shuf -n 1)"; echo "Random file: $$random_file"; \
+	curl -s -H 'Expect: ' -H "Content-Type: application/octet-stream" --data-binary "@$$dir/$$random_file" "${HOST}:${PORT5}/ekf" --output "out-ekf-$$random_file"
+#############################################################################################################################
+
+client-fib-once:
+	echo 30 | http ${HOST}:${PORT6}/fib
+# http ${HOST}:${PORT6}/fib?30
+
+client-fib-loadtest:
+	loadtest -n 10 -c 10 -P 30 "http://${HOST}:${PORT6}/fib" 
+
+client-fib-hey:
+	hey ${HEY_OPTS} -z 10s -c 72 -t 0 -o csv -m POST -d "30\n" "http://${HOST}:${PORT6}/fib"
+
+client-fib-wrk:
+	wrk -t 1 -c 1 -d 5s -R 1 "http://${HOST}:${PORT6}/fib?30"
+
+
+client-admin:
+	echo 5 | http ${HOST}:55555/admin
+
+client-terminator:
+	echo 5 | http ${HOST}:55555/terminator

tests/mt-emil/gocr_generate_dataset.sh

@@ -0,0 +1,13 @@
+#!/bin/bash
+
+VARYING=(72 108 144)
+
+for var in "${VARYING[@]}"; do
+    mkdir -p "$var"dpi
+    mkdir -p "$var"dpi-orig
+    for ((i=5; i<=15; i++)); do
+        shuf -n10 /usr/share/dict/american-english > "$var"dpi-orig/"$i"words.txt
+        pango-view --dpi="$var" --font=mono -qo "$var"dpi-orig/"$var"dpi_"$i"words.png "$var"dpi-orig/"$i"words.txt
+        pngtopnm "$var"dpi-orig/"$var"dpi_"$i"words.png > "$var"dpi/"$var"dpi_"$i"words.pnm
+    done
+done