Linux 5.7 To Improve Spreading Of Utilization, Other Scheduler Work
More improvements were queued in recent days to sched/core of CPU scheduler improvements on the table for the forthcoming Linux 5.7 kernel cycle.
One of the main patches to land was the work we talked about earlier this month for improved/faster spreading of CPU utilization. The change in the load balancing code is about ensuring there are pending tasks to pull otherwise the load balance will fail and further delay the spreading of system load. With the change now queued in sched/core, Linaro developer Vincent Guittot found that the average time for sysbench dropped slightly but the average peak time was a great deal less going from 21ms to 10ms while the absolute peak of 41ms to 21ms.
There is also optimized switching tasks inside shared cgroups, distributing tasks within affinity masks, and various other enhancements to the Linux scheduler code coming with v5.7.