An Early Performance Regression Hitting Highly Threaded Workloads On Linux 6.14-rc1

Written by Michael Larabel in Software on 4 February 2025 at 03:20 PM EST. Page 2 of 2. 14 Comments.

In other highly threaded workloads like the srsRAN 5G software, Linux 6.14-rc1 was again showing very severe regressions but if running only a single core/instance, the performance was unchanged.

In some of the threaded workloads there was less of a performance regression on Linux 6.14-rc1 but still a clear difference over Linux 6.12/6.13.

It wasn't all doom and gloom as in some threaded workloads like PostgreSQL or Memcached there were small improvements with Linux 6.14.

That's as far as I've pursued the regression(s) so far for the moment. With having a long backlog of other benchmarks/articles to work on with only so much hardware around and battling on a daily basis for survival among these difficult times for the web/ad industry and ad-blockers, I haven't dived any further into this performance regression. If it doesn't end up getting naturally resolved upstream in the coming days/weeks, I will dig further into it as time allows. But for now it does appear Linux 6.14 Git having some strange performance issues particularly for highly threaded / multi-core applications. It's not too surprising though given the early stage of the Linux 6.14 cycle and some of the changes being known at risk for performance regressions as possible culprits. There is a decent likelihood that the regression was introduced toward the end of the Linux 6.14 merge window (potentially even the link cited) as I did run some Linux 6.14 Git tests mid-way through the merge window initially with some of the same benchmarks but at that time hadn't seen any regressions -- though that run was on a completely different high core count AMD EPYC server.

UPDATE (5 Feb): Bisecting The Linux 6.14 Performance Regression With System76 Thelio + AMD Threadripper

14 Comments