Some Of The Workloads Still Seeing Lower Performance On Linux 5.5 Git

Last night I shared the results from what's causing one of the performance regressions in Linux 5.5 but sadly more regressions remain that are currently being tracked down.

Later today I hope to have the results to publish on a bisect of a second regression in Linux 5.5 Git. But overnight I did complete a run to rule out the workloads still affected even when disabling the kernel's AppArmor support per yesterday's article. These tests were done on the dual socket Xeon Platinum 8280 Cascade Lake server on Linux 5.5 Git as of yesterday.


So while the Hackbench scheduler benchmark results are back in good shape after nailing down that AppArmor regression outlined yesterday, other slowdowns on Linux 5.5 are still present:


The Pennant hydrodynamics application is one of the workloads surprisingly regressing on Linux 5.5, possibly due to the scheduler changes, but should know more shortly...


PostgreSQL in some configurations is also still slower than Linux 5.4 stable but for some configurations on the other hand seemed to be affected by the AA regression.

Cassandra is also still slower.

And some Memcached results are still showing hits to the performance on Linux 5.5.

It looks like I'll be spending my New Year's Eve doing more bisecting on the Linux kernel... Stay tuned.