AMD EPYC 7003 Series Performance Across Autumn 2021 Linux Distributions

[phoronix]

Phoronix: AMD EPYC 7003 Series Performance Across Autumn 2021 Linux Distributions

Recently I ran benchmarks showing how Ubuntu 21.10 performance has improved for AMD EPYC 7003 "Milan" even compared to just six months ago with Ubuntu 21.04. In this article is a broader look at AMD EPYC Milan on the autumn 2021 Linux distributions with firing up not only Ubuntu 21.10 but Fedora Server 35, Clear Linux 35150, CentOS Stream, and AlmaLinux 8.4 as other common alternatives in the Linux server space.

https://www.phoronix.com/vr.php?view=30677

[RedEyed]

It's hard for me to compare these benchmarks as zstd versions are different: v1.4 vs v1.5 knowing that v1.5 is much faster than v1.4 by design.

[Michael]

It's hard for me to compare these benchmarks as zstd versions are different: v1.4 vs v1.5 knowing that v1.5 is much faster than v1.4 by design.

In those benchmarks it was using system compress-zstd test profile that is using the Zstd as shipped by each of the distros.

[onlyLinuxLuvUBack]

Please could you add in opensuse leap 15.3 free ?

[RedEyed]

In those benchmarks it was using system compress-zstd test profile that is using the Zstd as shipped by each of the distros.

Yeah, got it.

Benchmark "Compression Level: 3 - Compression Speed" wonders me. CentOS with zstd v1.4.4 is 20% faster than Clear Linux with zstd v1.5.0, is it something related to IO/File System?

[igxqrrl]

Naive question: What tends to be the cause of the performance differences? Are different distributions tweaking kernel parameters?

I work for a small company that pays lots of money for hardware and for software licenses, but can't afford a dedicated expert to optimize our OS configurations (yes, penny-wise, pound-foolish; not my call). Further our software is only supported on one or two different OSes (usually CentOS7 and SUSE). It would be intriguing to try running in a CentOS container under, for example, ClearLinux.

[hoohoo]

I have always shied away from Clear because it has a quite different package and deploy model than every other distro I have ever used. But maybe it is time to bite the bullet.

[numacross]

Naive question: What tends to be the cause of the performance differences? Are different distributions tweaking kernel parameters?

I work for a small company that pays lots of money for hardware and for software licenses, but can't afford a dedicated expert to optimize our OS configurations (yes, penny-wise, pound-foolish; not my call). Further our software is only supported on one or two different OSes (usually CentOS7 and SUSE). It would be intriguing to try running in a CentOS container under, for example, ClearLinux.

Different kernel versions and configurations, compilers and their configuration, libraries and general software differences - would be a very short and shallow answer. It is a complex topic.

A good place to start would be reading Clear Linux docs on performance and related articles listed in it.

[s_j_newbury]

It would be interesting to see if any gains would be made with a Clear Linux distro rebuilt targetting Zen3 specifically.

[sdack]

It's hard for me to compare these benchmarks as zstd versions are different: v1.4 vs v1.5 knowing that v1.5 is much faster than v1.4 by design.

It is part of the point of these benchmarks. A distro's performance is obviously not only defined by compiler switches.

On the other hand should one not judge distros by only their performance. The "bells and whistles" of a distro are often more important than its raw performance.

I do miss Debian in these benchmarks for this reason because many distros are based on it and share many of Debian's features. Knowing how well these perform when compared to their parent does make these benchmarks more interesting. However, comparing Clear Linux to Ubuntu is like comparing apples to oranges, and then should include Gentoo as well. Until then are these benchmarks of distros on some specific CPU really just a half-hearted number toss.