Re: Performance regression: thread wakeup time (latency) increased up to 3x

From: Norbert
Date: Mon Nov 01 2021 - 21:35:09 EST



On 10/23/21 22:09, Norbert wrote:
(The more important part being SCHED_FIFO.)

Please ignore this side-comment. The decisive factor is isolated/nohz_full vs normal.

There are indications that "normal" times have become worse lately (measured in 5.14.14), relative to earlier, but in that area the fluctuations I see in my current tests are too large to get a clear picture.

Best, Norbert