Linux 5.0 Kernel Performance Is Sliding In The Wrong Direction

Rather than running with fast Core i9 or EPYC 2P boxes, for the last of the numbers to pass along today were from an Intel Core i5 8400 Coffeelake system. Due to time constraints and with the hardware being slower, I just benchmarked Linux 4.19 / 4.20 / 5.0 for now.

When looking at the geometric mean of all these tests, Linux 5.0 Git again came in slower than Linux 4.20. Though on this Core i5 system it was "just" 8% rather than ~10%.

In many of the same tests like BlogBench, Sockperf, and Apache Siege, this Core i5 8400 system is pointing towards lower performance in some areas with Linux 5.0.

I'll be running more tests through the weekend on more hardware and at least now having a few tests like Apache (with load via Siege), Sockperf, PostgreSQL, and others where they appear to be measurably slower on the new kernel for bisecting the issue to gain more insight on the Linux 5.0 performance. But overall there doesn't appear to be any very major, focused regression like we've seen in some of the past kernel series as a result of introducing new Spectre/Meltdown mitigations and other changes. Though with many of these apparent slowdowns being quite subtle could explain why the slower Linux 5.0 kernel performance has yet to be rectified unless it was another case like Linux 4.20 with STIBP where its original performance drop was sort of intentional albeit not properly communicated. Stay tuned for more tests.