Linux Workstation/Server Distribution Benchmarks For Winter 2016

[phoronix]

Phoronix: Linux Workstation/Server Distribution Benchmarks For Winter 2016

The latest for your enjoyment of our year-end comparison articles and benchmarks is a fresh comparison of various workstation/enterprise/server oriented Linux distributions when looking at relevant workloads. Testing for this distribution comparison being done from a Core i7 6800K Broadwell-E system while a desktop-focused Linux desktop comparison for winter 2016 will be posted still before year's end.

http://www.phoronix.com/vr.php?view=23904

[GreenByte]

So overall, clear Linux is the clear winner for the out of the box performance. That is no doubt since it was made by Intel, for Intel. But next year Zen/Ryzen should shake things up since it will be well performing and Intel will not really be able to fine tune it for Zen (nor they should want to).

[Michael]

So overall, clear Linux is the clear winner for the out of the box performance. That is no doubt since it was made by Intel, for Intel. But next year Zen/Ryzen should shake things up since it will be well performing and Intel will not really be able to fine tune it for Zen (nor they should want to).

Well for what it's worth many of the optimizations would carry over, when it comes to making more packages make use of AVX/AVX2, FMV, the more aggressive compiler flag defaults, etc. At least now with Zen there is AVX2 support and some other instructions in playing catchup with Intel CPUs. Clear Linux does CPUs back to ~Westmere era on Intel. But I haven't dived into the Kconfig used by the Clear Linux kernel to know whether they explicitly disable AMD chipset support or the like to know if Clear would boot on AMD hardware, but when Ryzen is out maybe I'll give it a shot and see if it works and how well.

[lucasbekker]

Very interesting to see these results, but they don't really help me select a workstation distro...
Of course, it's nice to have the fastest machine possible, but a WORKstation should focus on productivity. Choosing a rolling release or distro with the very latest compiler and packages destroys my productivity, because a lot of things don't work out-of-the-box. The time that I spend on fixing those things really makes the performance advantages moot.

[AsuMagic]

I've read somewhere that Ryzen CPUs should benefit from intel-specific optimizations (timing side, probably).

[YYYY]

Yo Michael:
" For this workstation/server-oriented distribution benchmarking, tested for this comparison was Ubuntu 16.04, Ubuntu 16.10, CentOS 7 1611, openSUSE Leap 42.3, Debian GNU/Linux 6.8, and Intel Clear Linux 12210. "

Must be openSUSE Leap 42.2 ;-)

[arjan_intel]

Very interesting to see these results, but they don't really help me select a workstation distro...
Of course, it's nice to have the fastest machine possible, but a WORKstation should focus on productivity. Choosing a rolling release or distro with the very latest compiler and packages destroys my productivity, because a lot of things don't work out-of-the-box. The time that I spend on fixing those things really makes the performance advantages moot.

Hmm I don't get how "rolling release" and "does not work out of the box" are correlated. There's good rolling release distros, and there's sloppy non-rolling ones....

[lucasbekker]

Well, sloppy non rolling distros are aplenty, but my experiences with rolling distributions and MATLAB for example are not great. It could be argued that many of those type of incompatibilities are not really the distributions fault, but none the less wreak havoc with my productivity.

[nicholas]

To question the validity of your tests, I assume some of the disk write tests could be considered thrashing. It is well known to set directories for such applications as nodatacow (see mariadb recommendations for example). So my question is - does blindly using the default BTRFS setup not give a mistaken and unfair comparison for opensuse?

[arjan_intel]

To question the validity of your tests, I assume some of the disk write tests could be considered thrashing. It is well known to set directories for such applications as nodatacow (see mariadb recommendations for example). So my question is - does blindly using the default BTRFS setup not give a mistaken and unfair comparison for opensuse?

to counter that; Michael measures out of the box, and shouldn't opensuse set up the database data directories correctly by default?
And nodatacow certainly also has integrity sacrifices, so it's not as easy as "just tune each distro to the max away from defaults to be fair".. once you start tuning as part of running the benchmarks, you end up on a slippery slope