Tag Archives: A Look At The Linux Kernel Performance From 4.10 To 4.20 Benchmarks

A Look At The Linux Kernel Performance From 4.10 To 4.20 Benchmarks

Here is a look at how the Linux kernel performance has evolved since Linux 4.10, which was released back in February of 2017, up through the current Linux 4.20 development cycle ahead of its debut at the end of December or early January. All of the Linux kernel benchmarks were done on the same venerable Intel Core i7 5960X system.

The Intel Core i7 5960X running at stock speeds was running with the ASRock X99 Extreme3 motherboard with its latest firmware, 16GB memory, and AMD FirePro V7900 graphics. Tests were done while running Ubuntu 16.04.5 LTS to avoid any packaging/systemd problems in running the older Linux kernel releases. Linux 4.10 through 4.19 were obtained via the Ubuntu Mainline Kernel PPA using each series’ x.y.0 kernel release. For Linux 4.20 it was built from source against the STIBP V2 fixes for those patches that will soon be queued to mainline that correct the overhead of that security feature originally introduced as part of the 4.20 merge window. So these Linux 4.20 benchmarks should be fairly representative of the performance of 4.20 final unless any other big security kernel changes land besides that improved STIBP/IBPB code.

Ubuntu 16.04 was running with its stock GCC 5.4 compiler and EXT4 file-system using all of the default options/settings. Via the Phoronix Test Suite a variety of benchmarks were run while also monitoring the AC system power consumption as an additional data point. If you enjoy our daily Linux benchmarking, consider taking advantage of our Cyber Monday special while in its final hours.

The time needed to do a defconfig build of the Linux x86_64 kernel (using the same 4.18 snapshot) hasn’t varied all that much as one of the tests looking at whether the newer kernels had any impact on code compilation performance. While Spectre/Meltdown mitigations have affected code compilation speed for some workloads, the kernel build test didn’t change all that much on this Core i7 5960X comparison.

Since Linux 4.13, the Blender 3D modeling software is running a few seconds slower than on 4.10 through 4.12.

The compile performance in CompileBench is actually up since Linux 4.12 and with 4.20 isn’t too much lower than it was at that high point.

In the read compiled tree test, the performance picked up in Linux 4.12 but then dropped off since with Linux 4.15… Why is that? Linux 4.15.0 is the first major kernel release that added initial Spectre/Meltdown work (not counting stable back-ports to point releases). But even with that drop off on post-4.14, the performance is about the same as two years ago on Linux 4.10.

Random writes with EXT4 were at their low point with Linux 4.18, but fortunately with Linux 4.19~4.20 the performance isn’t that far behind the fastest performance found on 4.17 followed by 4.13.

The FS-Mark performance is actually looking good on Linux 4.20 compared to the performance on the older kernels and a step-up compared to the current 4.19 stable.