Linux 5.9 Continues Working On A Variety Of Scheduler Improvements
For the all important Linux kernel scheduling code among the changes with the 5.9 cycle include:
- A new control for adjusting the default boost value for real-time workloads. This improvement is intended for better RT performance/efficiency particularly for Arm big.LITTLE systems but other platforms as well, particularly those battery powered where you might not want to be too aggressive with the default boosting behavior. The tunable is exposed via sysctl with "sched_util_clamp_min_rt_default."
- Frequency invariant scheduling is in better shape thanks to a few fixes/improvements for this feature in the x86 scheduling code by SUSE and Intel. This is about making better decisions on CPU usage by the scheduler based on today's modern CPUs and how their frequency scaling behaves along with turbo states.
- The deadline scheduler is now capacity-aware and has seen other improvements too.
- UClamp performance improvements for this utilization clamping functionality.
- Cleanups to the energy/power-aware scheduling.
These changes and more are outlined via this pull request.
Add A Comment
