Currently QA runs the performance scripts on same HW but different OS-es. Details here: https://wiki.yoctoproject.org/wiki/Performance_Test All system/distro settings are distro's defaults, that includes: - FS mount options: ext4 rw,relatime,data=ordered - IO scheduler: cfq and CPU governer: ondemand HW config is: SandyBridge i7-2600 CPU @ 3.40GHz HT-enabled (4 cores/8 threads), 8GB RAM, 1 x 500 GB HDD We have to investigate why there are big differences between them.
There still are performance differences between our Ubuntu and Fedora tests. A part of this was found to be python being compiled with profile guided optimisation which makes a big difference. Our buildtools tarball now has that.
We're still seeing a non-trivial difference between Fedora and Ubuntu.
The build performance difference between Debian-12 and Alma-8 is not huge and is now acceptable. Measurement Debian Alma Percent Diff Note sato build 1:14:59 1:15:13 + 0.31 % sato sstate 1:09.4 1:07.5 - 2.74 % kernel build 4:20.3 4:06.3 -12.1 % bitbake -p (1) 12.2 13.8 +13 % python version? https://valkyrie.yocto.io/pub/non-release/20250619-53/testresults/buildperf-debian11/perf-debian12-vk_master_20250619090025_7733ddf733.html https://valkyrie.yocto.io/pub/non-release/20250619-55/testresults/buildperf-alma8/perf-alma8-vk_master_20250619090031_7733ddf733.html I may look into the kernel build time but I suspect there is more variance than the ~1 second stddev reported on the build time graphs.
Since we didn't change anything, I've changed the resolution status to WORKSFORME