How Three BSD Operating Systems Compare To Ten Linux Distributions

[phoronix]

Phoronix: How Three BSD Operating Systems Compare To Ten Linux Distributions

Earlier this week I posted the results of a 10-way Linux distribution battle on the same Intel Xeon system and using all of the popular and latest Linux distribution releases. Taking things further, the article today has those results complemented by results on the Xeon system for several BSD operating systems. For seeing how the BSD performance stacks up to Linux, DragonFlyBSD, OpenBSD, and the FreeBSD-based PC-BSD were benchmarked.

http://www.phoronix.com/vr.php?view=22703

[TeamBlackFox]

Where's FreeBSD in this? PC-BSD isn't a good test for FreeBSD because its a downstream distribution of it - you should have included a vanilla FreeBSD installation with a comparable setup to the others. Also, FreeBSD's UFS2 is totally different from OpenBSD so this tells us little in the way of filesystem performance.

And where the hell is NetBSD? That would have been interesting to say the least.

[Michael]

Where's FreeBSD in this? PC-BSD isn't a good test for FreeBSD because its a downstream distribution of it - you should have included a vanilla FreeBSD installation with a comparable setup to the others. Also, FreeBSD's UFS2 is totally different from OpenBSD so this tells us little in the way of filesystem performance.

And where the hell is NetBSD? That would have been interesting to say the least.

Sans file-system differences, past tests I've done haven't shown much difference between PC-BSD and FreeBSD when using the same kernel and compiler.

With NetBSD it was running into problems on the system and didn't have the time to investigate.

[Adarion]

Impressive comparison, though the appearance of the BSDs seemed a bit uneven. ZFS is clearly good for a couple of things. Ah, well, but to be fair one had to test a couple of file systems for each case on the Linux kernel side. Back at the times e.g. reiser 3.6. was unbeaten for systems with a lot of small files.

Any reason for not including GPU side tests? I saw a relatively low end GPU was used, but why not? Afair the *BSDs do have kernel modesetting my now and the license of the AMD driver parts should allow easy porting.

[Michael]

Impressive comparison, though the appearance of the BSDs seemed a bit uneven. ZFS is clearly good for a couple of things. Ah, well, but to be fair one had to test a couple of file systems for each case on the Linux kernel side. Back at the times e.g. reiser 3.6. was unbeaten for systems with a lot of small files.

Any reason for not including GPU side tests? I saw a relatively low end GPU was used, but why not? Afair the *BSDs do have kernel modesetting my now and the license of the AMD driver parts should allow easy porting.

Because this particular system was using a Radeon HD 4550 that doesn't even work under Linux kernel with Radeon DRM due to mode-setting issues, so even under Linux the support was disabled and thus unaccelerated.

[dimko]

And where the hell is NetBSD? That would have been interesting to say the least.

Where is my favourite beast of all times, Gentoo? It constantly beats everything thrown at it in most Michaels benchmarks. Including BSD

[Asariati]

Phoronix: How Three BSD Operating Systems Compare To Ten Linux Distributions

Earlier this week I posted the results of a 10-way Linux distribution battle on the same Intel Xeon system and using all of the popular and latest Linux distribution releases. Taking things further, the article today has those results complemented by results on the Xeon system for several BSD operating systems. For seeing how the BSD performance stacks up to Linux, DragonFlyBSD, OpenBSD, and the FreeBSD-based PC-BSD were benchmarked.

http://www.phoronix.com/vr.php?view=22703

Why is ClearLinux missing from the first SQLite "Default Test Directory" and then suddenly shows up in Compile Bench? There is also one other Linux distribution missing from that test. Why?
Same question for the "FLAC Audio Encoding v1.3.1 test? Does that mean one can't encode audio on ClearLinux? Oh wait, it does show up in the "Lame MP3 Encoding v3.99.3" test !?

Shouldn't discrepancies like this be mentioned somewhere in the text?


And I have no idea what this table on the first page is trying to tell me. I think the font size is a little too small:

https://imgur.com/SQxEpOZ

[darkcoder]

Where's FreeBSD in this? PC-BSD isn't a good test for FreeBSD because its a downstream distribution of it - you should have included a vanilla FreeBSD installation with a comparable setup to the others. Also, FreeBSD's UFS2 is totally different from OpenBSD so this tells us little in the way of filesystem performance.

And where the hell is NetBSD? That would have been interesting to say the least.

Not sure, but I think these days they are much likely the same. At the begining PC-BSD used to modify/recompile many of the stuff, but now they even do simultaneously releases with FreeBSD. I doubt by heavy customizing they will be able to do that.

Also these days they use the low memory footprint desktop Lumina instead of KDE.

But you are right about NetBSD. It should be there. But performance wise probably is somewhere between PC-BSD/FreeBSD and OpenBSD.

[kuuchan]

I found this comparison interesting, and would like to see more BSD/Linux comparisons.

[ryao]

Was ashift set correctly on the ZFS vdevs? To answer the obvious response. There is no default. The setting depends on what the hardware reports and if the hardware lies it is blindly trusted. Were the distributions using XFS configured to give it a 4K block size?

Lastly, if you can compile production binaries at the numbers in your "compile bench" benchmark, you would likely have numerous companies trying to partner with you. Compilation is a CPU bound task. Numbers from speeding up a simulation of the IO pattern among other similarly inane things that I am sure experts in other areas could identify are meaningless unless your goal is to hurt the OSS community by flooding it with users who are even more confused about performance than they were before reading your benchmarks.