6-Way Enterprise Focused Linux Distribution Comparison With An Intel Core i9, Dual Xeon Gold Systems

[phoronix]

Phoronix: 6-Way Enterprise Focused Linux Distribution Comparison With An Intel Core i9, Dual Xeon Gold Systems

Here's our latest Linux distribution comparison with this time looking at the out-of-the-box performance of six Linux distributions while running a range of enterprise/workstation-focused benchmarks while using two systems. One system is a high-end Core i9 7980XE desktop system and the other a Tyan 1U Xeon Scalable server with dual Xeon Gold 6138 processors.

http://www.phoronix.com/vr.php?view=25553

[EthraZa]

But why is Clear Linux always winning those benchmarks? What's the trade-off here?

[arjan_intel]

But why is Clear Linux always winning those benchmarks? What's the trade-off here?

Clear Linux uses a bit more diskspace since it ships 2 copies of some key libraries (normal and AVX2 optimized with glibc picking the best for each system at runtime)

[Michael]

But why is Clear Linux always winning those benchmarks? What's the trade-off here?

Among the optimizations covered in this article and elsewhere: various kernel patches, defaulting to performance governor always, PGO'ing, various other compiler optimizations, selectively building with GCC vs. Clang compilers depending upon the package/bundle, shipping AVX and AVX-512 optimized libraries when the system supports it, optimized CFLAGS/CXXFLAGS defaults, etc... arjan_intel can probably provide a better rundown or documentation of Clear Linux optimizations.

[andyprough]

As usual, the EXT4 file system has a performance advantage. It would be interesting to see results with all distros on the same file system, rather than running some on XFS and others on EXT4.

[arjan_intel]

As usual, the EXT4 file system has a performance advantage. It would be interesting to see results with all distros on the same file system, rather than running some on XFS and others on EXT4.

it's a slippery slope... you can benchmark distro defaults (what Michael does) or you can start tweaking each one away from the defaults... but there's no end to that in sight.

[andrei_me]

arjan_intel clear Linux have it's own package right? Like apt for Debian, Pacman for arch, rpm for Fedora, etc

So a possible trade-off would be no package available for a specific software? As Clear Linux is server focused

[andyprough]

it's a slippery slope... you can benchmark distro defaults (what Michael does) or you can start tweaking each one away from the defaults... but there's no end to that in sight.

Those aren't the distro defaults though. Not for some of them.

[rrohbeck]

I'd be interested in seeing this comparison for Ryzen, Threadripper and Epyc.

[dagger]

I would love to see a similar tests for container orientated distros.