9-Way April 2014 Linux Distribution Benchmarks

[chrisb]

Now, talking about compiling software yourself, I'm not talking about sick CCFLAGS, but the possibility to not compile stuff in you don't want.

USE flags can reduce bloat by eliminating dependencies that aren't going to be used by the user, but they are unlikely to make any difference when it comes to benchmarks that are highly dependent on the CPU, kernel, or video stack. Computational benchmarks test the CPU and kernel scheduler. 3D benchmarks test those plus the 3D stack. Disk benchmarks test the kernel and drive. The code in these benchmarks typically runs in a well defined loop, the kind of extra baggage added by USE flags is never even executed.

That's the reason why Gentoo is in fact faster. It's a mathematical certainty!

I don't think that means what you think it means.

[frign]

Almost

The code in these benchmarks typically runs in a well defined loop, the kind of extra baggage added by USE flags is never even executed.

I should've made clearer that I was talking about general speed-benefits. In regard to tweaking USE-flags, it definitely improves startup-times and given most software handles things with internal data-structures, which are often dependent on what you compile them to include, benchmarks in this direction can also yield better results.

Being a Gentoo-developer myself, I often had the pleasure to encounter these positive effects in comparison to using other distributions, which often lack this flexibility.
The latter is the main reason why imho Gentoo is one of the finest distributions to use, given you have the experience.

[Mike Frett]

14.04

Why is 14.04 so slow compared to 12.04?. It's not much of an "upgrade". If I went purely by the Benchmarks, I would move to Debian from 12.04.

[FourDMusic]

CFLAGS don't speed up that much! It's all about the USE-flags instead, which can bring around 5-10% improvement.
Of course, if you benchmark Gentoo against Ubuntu and include the same features (read: all available) into a given program to make it an equal run, there won't be much of a difference.

Totally concur with this. There are better reasons why Gentoo kicks so much ass. It really teaches you how to an OS is built from the ground up since you install each part going through the Gentoo Handbook (which is really well written). Additionally, because you install only the software you are interested in, it runs very lean. The USE flags also are used to only enable the features in the software that you want, adding to the leanness of the distro. CFLAGS is nice icing on the cake - makes you feel like you're giving your CPU a present by using all of its ISA.

While portage can be a bit of a bitch when installing a package (sometimes have to run three times, once to find out there are packages that need to be unmasked, second with the --autounmask-write flag, and finally to actually install), it's a solid package manager and overlays are very useful.

[deanjo]

Noticed that the tests were done on openSUSE using the -desktop flavor of the kernel instead of the more server oriented and configured kernel-default package which is typically used in a server role.

[frign]

Very nice!

Totally concur with this. There are better reasons why Gentoo kicks so much ass. It really teaches you how to an OS is built from the ground up since you install each part going through the Gentoo Handbook (which is really well written). Additionally, because you install only the software you are interested in, it runs very lean. The USE flags also are used to only enable the features in the software that you want, adding to the leanness of the distro. CFLAGS is nice icing on the cake - makes you feel like you're giving your CPU a present by using all of its ISA.

While portage can be a bit of a bitch when installing a package (sometimes have to run three times, once to find out there are packages that need to be unmasked, second with the --autounmask-write flag, and finally to actually install), it's a solid package manager and overlays are very useful.

Well put!
I still wonder why the FSF marks Gentoo as a "proprietary" distribution, given it's the only distribution I know allowing to set licenses globally and package-specific.
For instance, if you only want libre software on your computer, just add

ACCEPT_LICENSE="-* @FREE"

to your make.conf and you're good to go.
Exceptions are easily made and definitely easier to maintain than Debian's "main", "contrib" and "non-free", which only let you decide between fully FOSS and a "dirty" tree.

Moreover, setting up kernels is trivial and LAMP has never been easier to set up than with Gentoo. Having recommended Debian for servers for years, I've now started to recommend Gentoo instead.

[gens]

@Michael

please add slackware to the comparison next time, if it isn't much trouble
last time you included it it got some scores noticeably lower, i'd like to see if that is still the case

[akincer]

I find the intra-distro comparisons most interesting (i.e. 12.04 vs 14.04) to see how newer features and configurations are impacting raw performance on certain tasks in default configurations. I'm not terribly surprised that 14.04 is not faster in all cases than 12.04. Not all of the changes in the kernel or other packages are performance improvements I would guess. Probably a small percentage are. New features are sexy. Fixing older code less so.