Linux 5.0 Kernel Performance Is Sliding In The Wrong Direction

With the Linux 5.0 kernel performance approaching the finish line, the past few days I've been ramping up my tests of this new kernel in our benchmarking farm. Unfortunately, when looking at the results at a macro level it's pointing towards Linux 5.0 yielding lower performance than previous kernel releases.

I haven't spotted any workload yet yielding a catastrophic performance regression on Linux 5.0 compared to 4.20 and earlier, but in a lot of workloads the 5.0 kernel is running faintly slower than 4.20. At first I thought it was isolated to just a particular box or two, but as my testing has increased, it's happened on a variety of at least x86_64 Intel/AMD hardware tested so far.

For a wide look at the kernel performance I have been running dozens of different benchmarks on each system, with some of the boxes going as far back as Linux 4.12 or the better part of two years ago. I've also taken the geometric mean of all tests successfully run on all tested kernel combinations and that's where it's clear Linux 5.0 is running slower than previous kernels...

All kernels were from the Ubuntu Mainline Kernel PPA and kept to their stock settings, including for any Spectre/Meltdown/security settings. First up I was testing an Intel Core i9 7960X thanks to its 16 cores / 32 threads for quickly pounding through many different benchmarks.

When looking at the geometric mean across the dozens of tests carried out, Linux 5.0 Git is clearly slower than its predecessors. With this box going back to Linux 4.12, keep in mind between Linux 4.14 and 4.15 is when Spectre/Meltdown came to light and began introducing in-kernel mitigations like PTI and Retpolines. But Linux 5.0 Git is clearly coming in as slower, which has been a similar outcome on other systems tested.