This diff adds the option to do iowait boosting to temporarily increase the frequency of tasks that are waiting on IO. This can improve IO performance in some situations.
Test using fio for a two minute benchmark with iowait boosting enabled using schedutil frequency governor. The number of jobs was large to cause iowaits to occur.
The test results aren't super conclusive, but there are many factors such as the latency of the disk(s) that make this hard to test. I'll try to tests on some other disks/filesystems.
This idea is borrowed from this patchset, but a far more crude implementation.
This diff adds the option to do iowait boosting to temporarily increase the frequency of tasks that are waiting on IO. This can improve IO performance in some situations.
Test using
fio
for a two minute benchmark with iowait boosting enabled usingschedutil
frequency governor. The number of jobs was large to cause iowaits to occur.With
--iowait-boost
disabled usingschedutil
:Compared to CFS with
performance
governor:The test results aren't super conclusive, but there are many factors such as the latency of the disk(s) that make this hard to test. I'll try to tests on some other disks/filesystems.
This idea is borrowed from this patchset, but a far more crude implementation.