Page 1 of 3 123 LastLast
Results 1 to 10 of 52

Thread: Would Phoronix do Gentoo Linux versus Ubuntu Linux benchmarks?

Hybrid View

  1. #1

    Default Would Phoronix do Gentoo Linux versus Ubuntu Linux benchmarks?

    Linux Magazine recently did benchmarks of Gentoo Linux and Ubuntu Linux alongside one another, but practically everyone on the Gentoo forums agrees that the comparison is flawed:

    http://forums.gentoo.org/viewtopic-t...ighlight-.html

    In the past, Phoronix has done benchmarks of alpha releases of distributions (e.g. Ubuntu 10.04), but has had a long standing omission of benchmarks involving Gentoo Linux, despite Gentoo Linux being a mainstream distribution that has a great deal to offer to the Linux community in ways that other distributions cannot.

    Would Phoronix be so kind as to do its standard benchmarks in as well as benchmarks similar to those done by Linux Magazine with Gentoo users' concerns addressed in a Gentoo Linux versus Ubuntu Linux comparison? Phoronix does excellent Linux benchmarks and it would be wonderful to see a meaningful comparison between the two distributions. It also would be wonderful to see Phoronix rectify its long standing omission of Gentoo Linux in its benchmarks.

    Before anyone mentions Sabayon Linux, benchmarks of Sabayon Linux do not count as benchmarks of Gentoo Linux, in particular because installing Sabayon Linux ignores all of the customization that is done on a typical Gentoo Linux System. While running Gentoo Linux does not automatically translate into higher performance, that customization often yields higher performance, which is something that is Sabayon Linux does not have.

    In case Phoronix is willing to do benchmarks of Gentoo Linux, here are some instructions:

    Code:
    The typical Gentoo user runs Gentoo Linux with packages from Gentoo's testing tree (software that is stable upstream, but has not had the formality of being declared stable by Gentoo's package maintainers), so a proper comparison between Gentoo Linux and Ubuntu Linux would involve using packages from Gentoo's testing tree (for x86_64, this would involve running 'echo "ACCEPT_KEYWORDS=\"~amd64\"" >> /etc/make.conf') and then masking any major software newer than the software that Ubuntu Linux uses. That would involving doing 'echo ">=sys-kernel/gentoo-sources-2.6.32" >> /etc/portage/package.mask', among similar commands for X, GCC, etcetera when comparing Gentoo Linux to Ubuntu Linux 9.10, such that all of the major software versions are identical. A list can be found at wikipedia:
    
    http://en.wikipedia.org/wiki/List_of...ommon_programs
    
    The sys-kernel/gentoo-sources, sys-devel/gcc, x11-base/xorg-server and x11-drivers/nvidia-drivers are likely the only software on the system that would require this treatment.
    
    The kernel should be compiled for the system's architecture, based off a .config from www.kernel-seeds.org. This should accurately simulate the level of kernel customization that is done by the typical Gentoo user, as the owner of that site has taken most of the things Gentoo users do, did them by default for us and published instructions regarding the things that tend to vary from system to system.
    
    Lastly, the recommendations for Gentoo involve having all software on the system compiled with "CFLAGS=\"-O2 -march=native -pipe -fomit-frame-pointer\"" and "CXXFLAGS=\"${CFLAGS}\"" in /etc/make.conf, although you could vary this if you want to try testing different optimization levels in your benchmarks, like Linux Magazine did.

  2. #2
    Join Date
    Jul 2008
    Location
    Berlin, Germany
    Posts
    821

    Default

    Such a comparison would be of little value. The major reason for using Gentoo is not performance but rather its flexibility.

    Besides, a benchmark run would be very hard to make fair and relevant at the same time:
    If you adjust the package versions in Gentoo to what you find in Ubuntu, this would be unfair to Gentoo, as you are not using what the developers recommend you to use. The other way round is not feasible, even. So you would penalize Gentoo for its flexibility.
    If instead you simply go with the latest stable or latest unstable releases, you will only compare performance of different X servers / kernels etc., and as these versions are constantly changing, chances are that the results are already outdated by the time they are published.

    Add to this that installing Gentoo requires reading and careful following of documentation, and you can probably see why no such benchmark comparison has been posted yet.

  3. #3
    Join Date
    Aug 2007
    Location
    Europe
    Posts
    401

    Default

    Quote Originally Posted by Shining Arcanine View Post
    Linux Magazine recently did benchmarks of Gentoo Linux and Ubuntu Linux alongside one another, but practically everyone on the Gentoo forums agrees that the comparison is flawed
    Of course they did, all four of them, at the time of this writing!

    One of them referred to optimizations, but did not specify which (s)he preferred. Which do you prefer?

    Which optimizations should be done for Ubuntu, as you, after all, can compile your own package for Ubuntu (and Debian) too, using about the same optimizations as for Gentoo.

    All comparisons are flawed, and most comparisons are more flawed than 50% of them!

  4. #4
    Join Date
    Dec 2008
    Posts
    984

    Default

    Quote Originally Posted by sabriah View Post
    of course they did, all four of them
    lol!

    5678910

  5. #5

    Default

    Quote Originally Posted by sabriah View Post
    Of course they did, all four of them, at the time of this writing!

    One of them referred to optimizations, but did not specify which (s)he preferred. Which do you prefer?

    Which optimizations should be done for Ubuntu, as you, after all, can compile your own package for Ubuntu (and Debian) too, using about the same optimizations as for Gentoo.

    All comparisons are flawed, and most comparisons are more flawed than 50% of them!
    The Gentoo forums are much more popular than these forums and at the time of this post, that thread has 316 page views versus 104 for this thread. The reason so far people posted is because they have nothing to add, not because they disagree. From my experience, if someone at the Gentoo forums disagrees about something, they say something.

    As for Ubuntu, Ubuntu is not designed to have the same level of customization that Gentoo has, and its community actively discourages it. Practically every community discourages going outside the scope of their distribution's package manager, Gentoo included, but Ubuntu's package manager and package maintainers attempt to handle the modification of configuration files and selection of preprocessor directives for users, which is something that Gentoo's package manager does not do. Ubuntu should not have any optimizations done for it, unless you consider installing its SMP i686 kernel to be an optimization, because anything more than that is illegal as far as it as a distribution is concerned, and would qualify as a fork of the Ubuntu distribution rather than any actual instance of the Ubuntu distribution.

  6. #6

    Default

    By the way, I would like to add that I listed how a standard Gentoo machine should be configured, which included the various optimizations that would be done, in my first post.

  7. #7
    Join Date
    Aug 2007
    Location
    Europe
    Posts
    401

    Default

    Quote Originally Posted by Shining Arcanine View Post
    By the way, I would like to add that I listed how a standard Gentoo machine should be configured, which included the various optimizations that would be done, in my first post.
    My apologies. I was in a grumpy mood.

  8. #8
    Join Date
    Nov 2007
    Posts
    1,353

    Default

    The only thing I want to add here is that while gentoo is incredibly flexible during install time, and once installed usually runs great permanently, it is a real bitch upgrade.

    Heres why....

    AppA-1.023 was compiled against Libx-4.86 during install time. After an upgrade LibX-4.86 is updated to LibX-5.08. AppA-1.023 is still linked against LibX-4.86 and must be recompiled against LibX-5.08 Portage doesnt do this for you.. And without specifically knowing which programs are compiled against which libraries, this cant be done easily.

    There is a utility called revdep-rebuild, but it is completely fucking backwards. Instead of recompiling the application, it recompiles the library, which has already been compiled during the update and doesnt actually need it.

    As far as I can tell the only sure fire way to fix these types of problems on gentoo is to do an emerge -e system && emerge -e world This unfortunately takes the better part of 2 days on my computer, and recompiles everything whether it is needed or not.

  9. #9
    Join Date
    Oct 2007
    Location
    Under the bridge
    Posts
    2,126

    Default

    Quote Originally Posted by sabriah View Post
    Which optimizations should be done for Ubuntu, as you, after all, can compile your own package for Ubuntu (and Debian) too, using about the same optimizations as for Gentoo.
    You don't understand, the point is to make Gentoo win the benchmark, not vice versa! I mean, otherwise the benchmarks will just show how pointless it is to turn to Gentoo purely for performance reasons.

    I wonder why they didn't test 64bit though. Who in his right mind would use Gentoo on a 32bit (i.e. Atom) CPU anyway?

  10. #10

    Default

    Quote Originally Posted by BlackStar View Post
    You don't understand, the point is to make Gentoo win the benchmark, not vice versa! I mean, otherwise the benchmarks will just show how pointless it is to turn to Gentoo purely for performance reasons.

    I wonder why they didn't test 64bit though. Who in his right mind would use Gentoo on a 32bit (i.e. Atom) CPU anyway?
    In a fair comparison (all important software versions being the same), Gentoo likely would win in benchmarks. It does not derive its name from the fastest species of penguin in the world for nothing. In principle, Gentoo could represent an upper limit for performance, which is an upper limit that other distributions's package maintainers might possibly be able to make their distributions closer to attaining.

    On the other hand, it is possible that Gentoo would not outperform other distributions in some areas, which is where benchmarks would really become interesting, because it would likely highlight a performance regression in the GNU Compiler that negatively affects all Linux distributions.

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •