Stretch the time-space more aggressively (lat_prio^2) if the task's latency priority is low (i.e., high value). This prioritizes latency-critical tasks more and gives bounded delay to throughput-oriented tasks.
Avoid the active/overflow CPU mask check when all CPUs are using.
Tighten the competition window to avoid unnecessary tasks that are competing with each other in the timeline.
In result, it improves:
The communication-intensive benchmark (e.g., perf bench sched messaging)
Interactive tasks when heavy compilation + stress-ng -c
This PR contains three tunings for
It changes three things:
In result, it improves:
perf bench sched messaging
)stress-ng -c