Initial Benchmarks Of OpenBSD 6.4, DragonFlyBSD 5.3, FreeBSD vs. Linux

[Michael]

Michael, any trick to run phoronix test suite on OpenBSD? I can reach all sites with ping, but `phoronix-test-suite network-info` tells me on all "Can Reach" simple "NO".
BTW, I'm not behind the proxy and also phoronix seems to run fine inside CentOS7 which is running in VMM on top of the tested OpenBSD. So the problem is probably in OpenBSD's php and its inability to reach your hosts. How have you solved that? I'm on 6.4-current though.

On OpenBSD you need to enable url fopen setting from its php.ini file as it ships with that support disabled for its PHP package.

[kgardas]

I guess OpenBSD + SQLite works well since OpenBSD always syncs writes by default, which is what SQLite takes extra effort to do on the other platforms.

Not true! OpenBSD FFS syncs only meta-data write by default. Data write is async!

[kgardas]

On OpenBSD you need to enable url fopen setting from its php.ini file as it ships with that support disabled for its PHP package.

Thanks! This indeed helps!

[kgardas]

Michael, thanks for testing OpenBSD! I'm glad it passed in all tests. The slowness in some of those is caused by synchronous meta-data write. I've been surprised that even "read" test of BlogBench still writes a lot of data. I'm on rotating rusty WD black 2.5" drive and with softdep mounted dir I've been able to beat your read on NVMe more than twice. So yes, intensive meta-data write on OpenBSD sucks. Good thing is that this is solvable by softdep (or hopefully in some future by WAPBL).

[vladpetric]

My experience with Clang/LLVM is that it produces significantly subpar code for x86-64, compared to gcc (clang/llvm was primarily meant for ARM, where it does a better job than gcc, IMO).

I'm willing to bet that some of the BSD benchmarks would benefit considerably from being built with gcc.

[aht0]

I'm willing to bet that some of the BSD benchmarks would benefit considerably from being built with gcc.

NetBSD and DragonFly have GCC for system compiler. Theory should be easy enough to check.,

[ren101]

Just be aware that OpenBSD 6.4, disables hyperthreading by default, so the CPU comparisons might be very well flawed.

[dillon]

Not to point out the obvious, but most of these benchmarks aren't testing the operating system at all... so you have to wonder why the results are so different. 7zip ? Really? What exactly is the operating system supposed to be doing there? It's a little cpu-bound program. Most of the other benchmarks are the same. There is clearly something very basic going on that is causing these massive differences in supposed performance. Michael, your time would be better spent actually figuring that out instead of just blindly posting the results and washing your hands of any responsibility.

-Matt