|
|
|
@ -89,6 +89,14 @@ generate_spec() {
|
|
|
|
|
for workload in "${workloads[@]}"; do
|
|
|
|
|
multiplier["$workload"]="$(get_random_from_interval $multiplier_interval_lower_bound $multiplier_interval_upper_bound)"
|
|
|
|
|
baseline_execution["$workload"]="$(get_baseline_execution "$results_directory" "$workload" $percentile)"
|
|
|
|
|
[[ -z "${baseline_execution[$workload]}" ]] && {
|
|
|
|
|
panic "Failed to get baseline execution for $workload"
|
|
|
|
|
exit 1
|
|
|
|
|
}
|
|
|
|
|
[[ -z "${multiplier[$workload]}" ]] && {
|
|
|
|
|
panic "Failed to generate multiplier for $workload"
|
|
|
|
|
exit 1
|
|
|
|
|
}
|
|
|
|
|
relative_deadline["$workload"]="$(calculate_relative_deadline "${baseline_execution[$workload]}" "${multiplier[$workload]}")"
|
|
|
|
|
{
|
|
|
|
|
echo "$workload"
|
|
|
|
@ -132,7 +140,7 @@ process_results() {
|
|
|
|
|
local results_directory="$1"
|
|
|
|
|
|
|
|
|
|
for workload in "${workloads[@]}"; do
|
|
|
|
|
# Filter on 200s, subtract DNS time, convert from s to ns, and sort
|
|
|
|
|
# Filter on 200s, subtract DNS time, convert from s to us, and sort
|
|
|
|
|
awk -F, '$7 == 200 {print (($1 - $2) * 1000000)}' < "$results_directory/$workload/benchmark.csv" \
|
|
|
|
|
| sort -g > "$results_directory/$workload/response_times_sorted.csv"
|
|
|
|
|
done
|
|
|
|
|