Announcement

Collapse
No announcement yet.

Ubuntu 7.04 to 8.10 Benchmarks: Is Ubuntu Getting Slower?

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

  • #31
    Originally posted by deanjo View Post
    Personally, I don't think PTS in it's current form is suitable for benching distro version vs distro version. Since a distro is a culmination of pre-built packages, to do a A/B test items such as the encoding tests and such should utilize the distro's pre-builts instead of compiling them from source which may not have the same build flags.
    Then that introduces a another variable due to the fact that the packages themselves may be faster or slower between versions. Having packages that invariable means that when benchmarking the tools used are taken out of the picture.

    Basic scientific/testing principles are that for a given test you only want to change one variable, everything else should stay the same. If a Distro is a collection of packages what are the constants across distros that we can use to test?
    Last edited by _txf_; 10-27-2008, 08:56 PM.

    Comment


    • #32
      Originally posted by _txf_ View Post
      Then that introduces a another variable due to the fact that the packages themselves may be faster or slower between versions. Having packages that invariable means that when benchmarking the tools used are taken out of the picture.

      Basic scientific/testing principles are that for a given test you only want to change one variable, everything else should stay the same. If a Distro is a collection of packages what are the constants across distros that we can use to test?
      In this case though you are benchmarking a "Distro", a complete solution. If you want to bench kernels then you should use the same Distro and version and update the kernel on that distro.

      If you were to do a shootout between a 2007 Mustang and a 2008 Mustang, you don't take them to your speedshop first and replace the factory parts with aftermarkets. You run them stock, as shipped.
      Last edited by deanjo; 10-27-2008, 09:27 PM.

      Comment


      • #33
        Originally posted by ssam View Post
        the tests were done on a core duo. that is a dual core CPU, so it can and is doing multiprocessing.
        You are right. I thought I read that the Core Duo wasn't a true dual core processor some time ago. This still does not mean the benchmarks used, took advantage of multiprocessing that is available.

        Originally posted by Supported Hardware: Ubuntu 8.04
        Multiprocessor support — also called “symmetric multiprocessing” or SMP — is available for this architecture. However, the standard Ubuntu 8.04 kernel image does not support SMP. This should not prevent installation, since the standard, non-SMP kernel should boot on SMP systems; the kernel will simply use the first CPU.

        In order to take advantage of multiple processors, you'll have to replace the standard Ubuntu kernel. You can find a discussion of how to do this in the section called “Compiling a New Kernel”. At this time (kernel version 2.6.24) the way you enable SMP is to select “Symmetric multi-processing support” in the “Platform support” section of the kernel config.
        https://help.ubuntu.com/8.04/install...supported.html

        The article makes no mention if the benchmarks are taking advantage if multiprocessing is being used.

        Comment


        • #34
          Originally posted by ssam View Post
          the tests were done on a core duo. that is a dual core CPU, so it can and is doing multiprocessing.
          You are right. I thought I read somewhere that Core Duo was not a true multicore processor.

          However, the person who performed the benchmarks did not specify if the benchmarks being used are taking advantage of multiprocessors since that is not enabled by default by reading the installation notes of 8.04. I would imagine 8.10 is not enabled by default either.

          https://help.ubuntu.com/8.04/install...supported.html
          Originally posted by Ubuntu Supported Hardware
          Multiprocessor support — also called “symmetric multiprocessing” or SMP — is available for this architecture. However, the standard Ubuntu 8.04 kernel image does not support SMP. This should not prevent installation, since the standard, non-SMP kernel should boot on SMP systems; the kernel will simply use the first CPU.

          In order to take advantage of multiple processors, you'll have to replace the standard Ubuntu kernel. You can find a discussion of how to do this in the section called “Compiling a New Kernel”. At this time (kernel version 2.6.24) the way you enable SMP is to select “SMP support” in the “General setup” section of the kernel config.

          Comment


          • #35
            hey phoronix, you make benchmarks using Fglrx driver? LOL.
            why do you do benchmarks with alha drivers (because that's what they are)? Especially when you can do them with a mature driver, named Nvidia.

            -1 for this article.

            as of now, this article is FUD.
            Last edited by bulletxt; 10-27-2008, 11:34 PM.

            Comment


            • #36
              How about desktop effects?

              Do you by any chance have desktop effects on? I know for sure 6.06 doesn't have them, while 8.04 and 8.10 do. I don't recall if 7.10 had them on by default. ATI driver traditionally didn't support the Composite extension but now does.

              So, the X1400 is listed as supporting "Turbocache" which lets the card use system memory. I'd think enough textures for a desktop should fit on the card, but if the driver putting these in system memory, compositing will chew through your memory bandwidth.

              Alternately, (but still blaming desktop effects) perhaps Composite's just a drag with the ATI driver even if everything is on the card's RAM.

              Either way, if they're on I'd certainly test with the effects off and see if it makes a difference.

              (I expect large fluctuations in the compiling benchmarks between Ubuntu versions, since there's probably 4 different versions of gcc in play.)

              Comment


              • #37
                Originally posted by hwertz View Post
                Do you by any chance have desktop effects on? I know for sure 6.06 doesn't have them, while 8.04 and 8.10 do. I don't recall if 7.10 had them on by default. ATI driver traditionally didn't support the Composite extension but now does.

                So, the X1400 is listed as supporting "Turbocache" which lets the card use system memory. I'd think enough textures for a desktop should fit on the card, but if the driver putting these in system memory, compositing will chew through your memory bandwidth.

                Alternately, (but still blaming desktop effects) perhaps Composite's just a drag with the ATI driver even if everything is on the card's RAM.

                Either way, if they're on I'd certainly test with the effects off and see if it makes a difference.

                (I expect large fluctuations in the compiling benchmarks between Ubuntu versions, since there's probably 4 different versions of gcc in play.)
                From the article:

                With the Mobility Radeon X1400 we had used the ATI Catalyst 8.10 driver from AMD's web-site except for when using Ubuntu 8.10 where we had used the Catalyst 8.10 (fglrx 8.54) driver that ships with Ubuntu as it has Linux 2.6.27 and X.Org 7.4 support. Compiz and Enhanced Intel SpeedStep Technology (EIST) were disabled during testing.

                Comment


                • #38
                  I have noticed the same. But since Gutsy the system is not useable on heavy io. It's not an issue of the scheduler, because CentOS or Feisty with the cfq on the same machine works fine. Disabling the Fair Group Scheduler brings only marginal changes. In my test other distributions have the same issue, but not as awful as Ubuntu has.

                  I think it is a kernel issues since 2.6.20. The RHEL 2.6.18 works fine. I have tested the kernel even with Fedora 9. The issue is not related on SATA/PATA, 32-/64-Bit or 1 Core/2 Cores.

                  Currently Intrepid is really slow. I am thinking about switching back to windows, because I am working with this issue for more than a year. And the performance is awful.

                  I have tested the desktop responsiveness on heavy io. With Intrepid I only need to update more than three packages and gets desktop freezes for more than 20 seconds.

                  If someone has an hint for me, how to solve or how to reduce the problem. Please tell me.
                  Last edited by ebird; 10-28-2008, 07:02 AM.

                  Comment


                  • #39
                    Originally posted by hwertz View Post
                    Do you by any chance have desktop effects on?
                    As said in the article, Compiz was disabled.
                    Michael Larabel
                    http://www.michaellarabel.com/

                    Comment


                    • #40
                      Borked!

                      Hi there,

                      Last night i runned the test on my Dell Latitude D505 (P-Mobile 1.7GHz, 2GB RAM, 160 GB IDE-HDD) and here are my scores (In brackets are the scores for the tested T60). I run Ubuntu 8.10 on my system :

                      Bandwidth (sequential read) : 1720.74 MB/s (1838.94 MB/s)

                      LAME-MP3 : 73.60s (120.83s)

                      OGG : 45.53s (69.81s)

                      FLAC : 33.13s (59.19s)

                      WAVPACK : 41.03s (68.20s)

                      FFMPEG : 41.03s (70.09s)

                      SQLITE : 23.45s (23.06s)

                      SCIMARK (composite) : 265.47 Mflops (187,83 Mflops)

                      Tandem-XML (write) : 59.54s (104.22s)

                      My CPU is a little bit slower than the one of the T60, so each of the score should be a little bit more lower than the T60 ones. But nearly all of them are significantly better!

                      Is it possible to rerun the tests on the T60 and disable the tracker-indexing service before running the tests? I think this might be the cause of the really bad scores with Ubuntu 8.10.

                      Comment


                      • #41
                        Where all these tests run from the same part of the harddrive? Or did the distros have their own partitions? If so, Feisty would be faster if it was in the first partition, as disk access is quicker nearer to the centre of the disk.

                        Also this previous Phoronix article has different things to say about the performance between Ubuntu 7.10 and Ubuntu 8.04 (especially lame):
                        http://www.phoronix.com/scan.php?pag..._710_804&num=1
                        Last edited by Xanikseo; 10-28-2008, 07:41 AM.

                        Comment


                        • #42
                          Originally posted by thacrazze View Post
                          The cause for the bad performance is the CFS (Completly Fair Scheduler), I think. It's standard since 2.6.23/24.

                          World of Warcraft on Wine lags terrible and freezes every some seconds since the introducing of CFS.

                          I hope OpenSolaris supports my hardware completly soon, so that I can change to this OS.
                          Slowaris? don't fool yourself.

                          Comment


                          • #43
                            I am just wondering does a dual core CPU need SMP like a dual CPU system need it?
                            I think it was the case but is it still the case?
                            Last edited by shreg; 10-28-2008, 12:06 PM.

                            Comment


                            • #44
                              yes it is. multicore is is the same as multi cpu. Just with a bit different packaging. If you have a bigger AMD system (like dual socket, quad core) you even need NUMA to take advantage of the hardware.

                              So without SMP support in the kernel one core would always lie dormant.

                              Comment


                              • #45
                                Originally posted by energyman View Post
                                So without SMP support in the kernel one core would always lie dormant.
                                It's probably more accurate to say 1 core is active given that there are quadcores and such now.

                                Comment

                                Working...
                                X