16-Way Linux OS Performance Comparison

[sireangelus]

As always, i see that gentoo does not get recompiled by changing the make.conf to -fomit-frame-poitenter and at least -march=native.. that's the "speed magic" that gentoo users claim over the other distros.

[rdnetto]

I think that's the point: There's no winner. All the distributions preform the same once you bother installing the proper firmwares and drivers. That's to say, when trying to compare different linux\bsd distributions, you should look at usability, ease of installation and maintenance, package availability, support and so on... Instead of performance.

Perhaps, but it seems to me that this comparison is more about comparing out of the box performance.
Also, there are performance-related aspects where distros do vary. For example, recently I compared the boot times of a few major distros (measured from GRUB to the display manager):

Debian: 20.1s
Kubuntu: 26.1s
Sabayon (OpenRC): 31.6s
Sabayon (OpenRC parallel): 29.8s
Sabayon (systemd): 26.8s

Given the difference between the best and worst case is 10 sec, or ~50%, that would certainly make a noticeable difference. (Although Debian is probably faster only because it just had KDE installed, with none of the useful services the others ship with.)
Something like this would be very useful to have in a multi-way comparison, since it would make a real difference to people.

I'm personally using Sabayon since Debian has become quite convoluted lately with all the half functioning user land packages. systemd is being developed too quickly for Debian to keep up so there's just too much breakages and rust collecting. Being familiar with Gentoo from the server, I figured a binary derivative should be fine. And it did.

I'm also a recent convert to Sabayon from Debian (switched just last week). Gentoo really does have some nice aspects (e.g. eselect), once get past/around the source-based packages. It's also a lot more stable than I expected - so far I've only had to deal with one bug (the screen locking issue in KDE 4.10).

[jumico]

It did show that Mageia lagged behind a little more than most distributions.

[ferry_th]

This article came from the future.

[serj]

Phoronix: 16-Way Linux OS Performance Comparison

Building on our earlier 11-Way Linux/BSD Platform Comparison, starting a new week we're up to a 16-Way Linux operating system comparison. Added in now are results from PCLinuxOS, ROSA, the lightweight antiX distribution, and then the Gentoo-based Sabayon and Calculate Linux Desktop distributions.

http://www.phoronix.com/vr.php?view=18809

Could you give the details for Mageia?
It's strange it's slower.
What kernel flavor was installed?
kernel-desktop? kernel-server?
"rpm -qf /boot/vmlinu*" should tell you

From your tests, it's not clear whether you were running 32 or 64bit.
If you were running in 32bit mode, then Mageia would have installed kernel-server in order to be able to manage the 8Gb of RAM, which might be slower and could explain the difference.

[Sergio]

I'm trully glad (and admiteadly surprised) to see how DragonFlyBSD, an OS with zero support whatsoever from big companies and only a handfull of developers, manage to compete (and, ocasionally, outperform) big OS's like Linux and FreeBSD. Good for them.

[leio]

Thanks for yet another completely useless and off the mark benchmark with completely bogus conclusions. All these benchmarks do are compare performance from different versions of kernel, gcc and co, with all benchmarks being PTS self-compiled tests instead of distribution ones, also almost all of them being mostly unaffected by the performance of dependent libraries from the distributions own packages. Also almost all of the conclusions that are actually done, are completely wrong. For instance:

"While some Linux distributions will do various customization work to the desktop and other areas of the system, it generally doesn't translate to any real-world performance improvements." - NO real world performance is being measured in ANY of the benchmarks; only PTS self-compiled specific benchmarks that mostly depend on toolchain and kernel, or which video driver happens to be picked, nothing which is commonly used in the real world, or nothing that would be used from the distribution as they are self-compiled.

"The Gentoo-based Calculate Linux Desktop and Sabayon Linux distributions didn't offer any compelling performance advantages over the non-Gentoo distributions, contrary to some Gentoo users believing it's a magical speed demon." - complete misunderstanding of why those Gentoo users would have such beliefs and what Sabayon and Calculate Linux actually are compared to Gentoo Linux itself; I'd expect some basic knowledge and understanding on what is being measured, what isn't and so on before such claims are made; but maybe such shoddy reporting just leads to good ad revenue. I would be perhaps theoretically willing to provide some stage4's when we can find some common hardware ground to provide a more ready system, but I don't think I'd feel like it for phoronix, as the resulting benchmarks would probably be measuring something completely wrong yet again.
Sadly I mostly hoped to find a good comparison of binary distributions out of this article, but due to aforemention paragraphs can't find any of that either. No measurements of Gentoo were expected, but such claims on Gentoo itself are outrageous.

All the other conclusions just reinforce the fact that this is not a distribution comparison at all, except the comparison of apples and oranges with the default video drivers (which simply reiterate on all the video driver benchmarks ran on this site).

As always, i see that gentoo does not get recompiled by changing the make.conf to -fomit-frame-poitenter and at least -march=native.. that's the "speed magic" that gentoo users claim over the other distros.

-fomit-frame-pointer should be included in -O2 these days on many architectures, because it doesn't disturb debugging thanks to location lists features added somewhere in gcc 4.0..4.2 and matching gdb era. It does however in many situations disturb some sample based profiling (e.g sysprof), so I in fact pass -fno-omit-frame-pointer these days

[talvik]

Thanks for yet another completely useless and off the mark benchmark with completely bogus conclusions.

Speak for yourself I want *REAL* performance benchmarks. I don't want custom compilers and flags.
If distro X choose some kernel/gcc/lib version that makes it perform better than Y. That's the performance you will experience when you use distro X.
If "distro benchmarks" used the same kernel, scheduler and compiler what's the point?

[sireangelus]

Speak for yourself I want *REAL* performance benchmarks. I don't want custom compilers and flags.
If distro X choose some kernel/gcc/lib version that makes it perform better than Y. That's the performance you will experience when you use distro X.
If "distro benchmarks" used the same kernel, scheduler and compiler what's the point?

custom flags is what gentoo is aobut... you can't do a performance comparison without it..