Announcement

Collapse
No announcement yet.

Intel Ultrabook: Fedora 18 vs. Ubuntu 13.04 Tests

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Intel Ultrabook: Fedora 18 vs. Ubuntu 13.04 Tests

    Phoronix: Intel Ultrabook: Fedora 18 vs. Ubuntu 13.04 Tests

    Our latest benchmarks from the ASUS S56CA-WH31 Intel Ultrabook is comparing the performance of this Ivy Bridge laptop between Fedora 18 and a recent Ubuntu 13.04 development snapshot under various workloads...

    http://www.phoronix.com/vr.php?view=MTM0MzI

  • #2
    What's with that crazy performance diff between Fedora and Ubuntu on Apache?

    Comment


    • #3
      Originally posted by Sonadow View Post
      What's with that crazy performance diff between Fedora and Ubuntu on Apache?
      I had that same question.... possibly different apache versions, and the one in fedora has a bug. Different default configurations FOR apache, SELinux on fedora may be getting in the way. If Rahul or any of the other RedHat / Fedora guys could weigh in, itd be appreciate because that is a serious regression in comparison

      Comment


      • #4
        Originally posted by Sonadow View Post
        What's with that crazy performance diff between Fedora and Ubuntu on Apache?
        It's been so for a while which is very strange because the other results is more even.
        My guess would be some strange compile-time options in Fedoras version.

        Comment


        • #5
          In Apache case, result does not tell about the configuration. By default, in Fedora, apache will run confined meaning some scripts will not be allowed to run for security purpose.

          Comment


          • #6
            I don't know why anybody would care about static web page performance. It's essentially a test to see who could be the fastest at doing nothing. Either of them are going to be plenty fast enough.

            There could be a lot of good reasons why Fedora would be slower at that. Default threading model, for example. Maybe Ubuntu has it setup so that more processes and threads are started by default or something like that.

            I'd look at the obvious differences first before I'd start speculating about compile options or selinux getting in the way. SElinux overhead should only be a few percentages at most.

            Comment


            • #7
              Originally posted by phoronix View Post
              Phoronix: Intel Ultrabook: Fedora 18 vs. Ubuntu 13.04 Tests

              Our latest benchmarks from the ASUS S56CA-WH31 Intel Ultrabook is comparing the performance of this Ivy Bridge laptop between Fedora 18 and a recent Ubuntu 13.04 development snapshot under various workloads...

              http://www.phoronix.com/vr.php?view=MTM0MzI
              So, we're testing yet to be released Ubuntu with the current release of Fedora?
              F19 has branched, why not use that?

              Comment


              • #8
                Originally posted by drag View Post
                I don't know why anybody would care about static web page performance. It's essentially a test to see who could be the fastest at doing nothing. Either of them are going to be plenty fast enough.

                There could be a lot of good reasons why Fedora would be slower at that. Default threading model, for example. Maybe Ubuntu has it setup so that more processes and threads are started by default or something like that.

                I'd look at the obvious differences first before I'd start speculating about compile options or selinux getting in the way. SElinux overhead should only be a few percentages at most.
                Yeah, I'd guess it's just not a realistic comparison - possibly Fedora uses the upstream defaults for Apache config, while Ubuntu ship a pre-tuned version. If so, it's a meaningless benchmark, since nobody is going to run a real server on the default configuration. It'd be more interesting to see a tuned vs tuned comparison, to show up any differences in the actual OS (like the other benchmarks are doing).

                Comment


                • #9
                  Originally posted by liam View Post
                  So, we're testing yet to be released Ubuntu with the current release of Fedora?
                  F19 has branched, why not use that?
                  Fedora builds before Beta are very slow due to the use of debug-enabled kernels (a few other packages have performance-impacting debug stuff enabled early in the build cycle too, IIRC).

                  These kinds of comparisons are usually of pretty limited value, though; the basic answer is that any major difference in performance between two mainstream Linux distributions is almost always the result of some kind of specific, transient bug that will get fixed, an error in the test, or some configuration setting (in which case it'd make much more sense just to tweak your configuration for whatever distro you wind up picking than pick your distro based simply on what it picks for a default configuration for some component or other). These kinds of comparisons are sometimes useful for turning up bugs, but that's about it. Linux distributions are ultimately built in much the same way from most of the same source code; it's just to be expected they'll usually perform about the same at the same workload, if you organize your test correctly. Some kind of idea of 'general performance' is not a good factor to consider in choosing a Linux distribution; it makes much more sense to choose your distribution based on the things that actually differ between distros, and then optimize performance of your particular workload once you've picked a distro.

                  Comment


                  • #10
                    Originally posted by AdamW View Post
                    Fedora builds before Beta are very slow due to the use of debug-enabled kernels (a few other packages have performance-impacting debug stuff enabled early in the build cycle too, IIRC).

                    These kinds of comparisons are usually of pretty limited value, though; the basic answer is that any major difference in performance between two mainstream Linux distributions is almost always the result of some kind of specific, transient bug that will get fixed, an error in the test, or some configuration setting (in which case it'd make much more sense just to tweak your configuration for whatever distro you wind up picking than pick your distro based simply on what it picks for a default configuration for some component or other). These kinds of comparisons are sometimes useful for turning up bugs, but that's about it. Linux distributions are ultimately built in much the same way from most of the same source code; it's just to be expected they'll usually perform about the same at the same workload, if you organize your test correctly. Some kind of idea of 'general performance' is not a good factor to consider in choosing a Linux distribution; it makes much more sense to choose your distribution based on the things that actually differ between distros, and then optimize performance of your particular workload once you've picked a distro.
                    I hesitate to say this, but for most users it is safer to stick to defaults than to try to tweak things.
                    For one thing, you need to know what tweaks are available, then you need to know what it is they do. Next you need to be able to reliably test these changes to see if they actually make a difference in what you care about (this later one is the biggie, IMHO).
                    I know you are aware of these things, but I wanted to explain why I think it is safer to stick with the defaults for most users, and why benchmarking can be useful, even if that which is done on phoronix isn't of the most useful kind.
                    Your point about debugging is fair, though.

                    Comment


                    • #11
                      Originally posted by liam View Post
                      I hesitate to say this, but for most users it is safer to stick to defaults than to try to tweak things.
                      For one thing, you need to know what tweaks are available, then you need to know what it is they do. Next you need to be able to reliably test these changes to see if they actually make a difference in what you care about (this later one is the biggie, IMHO).
                      I know you are aware of these things, but I wanted to explain why I think it is safer to stick with the defaults for most users, and why benchmarking can be useful, even if that which is done on phoronix isn't of the most useful kind.
                      Your point about debugging is fair, though.
                      Well, sure, I was assuming a certain level of knowledge. The corresponding advice for most 'end users' is "pick a distro based on whatever the latest groupthink is and then DON'T FREAKING FIDDLE WITH STUFF because smarter people than you have already set it all up properly. Really. Don't do what that kid on a forum thread is telling you to do. Stop. Put the text editor down."

                      But it usually takes people a few painful experiments to come to that conclusion. =)

                      Comment


                      • #12
                        Originally posted by AdamW View Post
                        Well, sure, I was assuming a certain level of knowledge. The corresponding advice for most 'end users' is "pick a distro based on whatever the latest groupthink is and then DON'T FREAKING FIDDLE WITH STUFF because smarter people than you have already set it all up properly. Really. Don't do what that kid on a forum thread is telling you to do. Stop. Put the text editor down."

                        But it usually takes people a few painful experiments to come to that conclusion. =)
                        You know, you're being glib, I think, but your comment is either a sad fact, or a data point suggesting more needs to be done to see what it is the "user" actually wants/needs.
                        I know, good luck with that

                        Something I don't understand about desktops distros, especially ones like ubuntu, but also, in theory, fedora, is why they aren't defaulting to a preemptive kernel like windows and osx. I understand that some decisions have been made in fedora that supporting more than one kernel is too many kernels to support, but it's not as though RH doesn't make extensive use of the rt patches so it would seem it would be useful to have more people testing them. That change, by itself, would stop a good amount of tinkering, I think.

                        Comment


                        • #13
                          Originally posted by liam View Post
                          I hesitate to say this, but for most users it is safer to stick to defaults than to try to tweak things.
                          For one thing, you need to know what tweaks are available, then you need to know what it is they do. Next you need to be able to reliably test these changes to see if they actually make a difference in what you care about (this later one is the biggie, IMHO).
                          I know you are aware of these things, but I wanted to explain why I think it is safer to stick with the defaults for most users, and why benchmarking can be useful, even if that which is done on phoronix isn't of the most useful kind.
                          Your point about debugging is fair, though.
                          Sure, but if you're running a web server and you're sensitive about the performance, *you* need to know these things. Good defaults are nice, but the people providing the default configuration don't know what hardware you're running, so there are limits to what they can do. If you care about server performance, it's your job to know what the settings mean, and run your own benchmarks until you're happy it's working optimally.

                          Comment


                          • #14
                            i don't like theses distros, suits well for desktop but not for servers, they are not enough stable

                            Comment


                            • #15
                              Originally posted by Delgarde View Post
                              Sure, but if you're running a web server and you're sensitive about the performance, *you* need to know these things. Good defaults are nice, but the people providing the default configuration don't know what hardware you're running, so there are limits to what they can do. If you care about server performance, it's your job to know what the settings mean, and run your own benchmarks until you're happy it's working optimally.
                              I'm not arguing against any of that. My only point was that defaults need to be well chosen for whatever arena the OS is intended.
                              Testing server functionality on fedora is a bit silly as few would run fedora as a server (or at least a production server).
                              What would be nice would be a few meta packages that "optimize" a distro for various uses. So, a desktop should install a preempt kernel with various knobs twisted as the packager deems appropriate (hopefully having tested said tweaks and found settings that "work well for most" on the intended arena), a server would keep the default cooperative kernel, and, perhaps, a few options the user could specify after install to indicate how the server is intended to be used (application, web server, file server, whatever), and a low latency desktop for content creators (both visual and audio content creators). These things are possible b/c there are distros that do exactly these things, but it seems a waste of resources to have completely separate distros that make only a few changes (changes which could be handled through meta-packaging). All of these should provide good starting places for the tweaker (hopefully with some docs as too what settings they have changed from the default of the server).
                              I hope this makes my point clearer.

                              Comment

                              Working...
                              X