The Linux 5.7 Scheduler Changes Bring Prominent Additions For Intel & Arm CPUs
Ingo Molnar on Monday sent in the scheduler updates for the Linux 5.7 kernel that saw its merge window open at the start of this week. For the Linux 5.7 cycle are a number of prominent scheduler additions.
Highlights on the scheduler side for Linux 5.7 include:
- NUMA scheduling updates so the load balancer and placement logic do not fight each other in order to improve locality and utilization with less migrations.
- Thermal pressure tracking for thermally-overloaded systems. This work focused initially on Arm CPUs is for ensuring better task placement on hot/overheating cores that in turn are being limited/down-clocked.
- Schedutil frequency invariance for supported Intel CPUs. This should yield better performance and power efficiency. This frequency invariance work should work on Skylake X and newer, Xeon Phi, and select Atom/Goldmont parts. This can yield some big wins and we'll be running some benchmarks internally in the near future. It's thanks to this big improvement that P-State is working to migrate to Schedutil from powersave by default.
- Improved capacity utilization for big.LITTLE platforms thanks to asymmetric CPU capacity wake-up scanning.
- Real-time scheduling fixes/improvements.
- Various other fixes and improvements.
The full list of scheduler alterations via this PR that was merged on Tuesday.
Highlights on the scheduler side for Linux 5.7 include:
- NUMA scheduling updates so the load balancer and placement logic do not fight each other in order to improve locality and utilization with less migrations.
- Thermal pressure tracking for thermally-overloaded systems. This work focused initially on Arm CPUs is for ensuring better task placement on hot/overheating cores that in turn are being limited/down-clocked.
- Schedutil frequency invariance for supported Intel CPUs. This should yield better performance and power efficiency. This frequency invariance work should work on Skylake X and newer, Xeon Phi, and select Atom/Goldmont parts. This can yield some big wins and we'll be running some benchmarks internally in the near future. It's thanks to this big improvement that P-State is working to migrate to Schedutil from powersave by default.
- Improved capacity utilization for big.LITTLE platforms thanks to asymmetric CPU capacity wake-up scanning.
- Real-time scheduling fixes/improvements.
- Various other fixes and improvements.
The full list of scheduler alterations via this PR that was merged on Tuesday.
Add A Comment