Running Linux 5.15-rc1 Causing A New Slowdown... Here's A Look

With the clear trend of build time tests going up on Linux 5.15 and given the real-world and widespread significance, it was on to bisecting...

Long story short, from the initial round of bisecting the regression appears attributed to the memcg changes this cycle. In particular, aa48e47e3906c332eaf1e5d7b58be11d3509ad9f appears to be the first bad commit, the change around infrastructure to flush memcg stats.

This didn't end up coming entirely as a big surprise... Earlier in the Linux 5.15 merge window, Linus reverted some of the other memcg patches for causing performance regressions but this particular commit in question remains left in the kernel as of writing. There was this message by Linus Torvalds where he did make mention of the aa48e47e3906 patch but that it would likely be a bigger deal to revert.

There hasn't seemingly been any discussion in more recent days around that patch and as of writing hasn't been reverted. I have let Linus know though about being able to confirm a regression there, so we'll see what happens for 5.15.

Given the commit, it's possible other workloads are also being negatively impacted on Linux 5.15. So far though it's been the common area of noticing build compilation times slowing, but will run some follow-up tests now with the many other benchmarks looking for other changes in 5.15 performance.

That's all for now. If you appreciate my constant Linux benchmarking, open-source benchmarking work, etc, please consider showing your support for this time intensive and costly operation. Join Phoronix Premium, make a PayPal tip, or at the very least do not use any ad-blocker if you wish for such content to continue. Thanks for your support and now on to looking at other Linux 5.15 performance changes.