Announcement

Collapse
No announcement yet.

Ubuntu 10.04 Is Off To A Poor Performance Start

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

  • #16
    Originally posted by movieman View Post
    Sure, but most people will just use the defaults; certainly I did when I installed 9.10 on my netbook, though only because there's nothing on there important enough that I'd cry if ext4 bugs ate it .
    There isn't really a bug... it's a feature. Hey that sounds familiar

    But all kidding aside... the Ext4 feature that we're talking about is a, I believe tripple, integrety check just to make the FS more safe. Adding another one of those Oracle integrety layers makes the system, of course, a lot slower

    Comment


    • #17
      Originally posted by garytr24 View Post
      Ubuntu doesn't follow this development model. It imports packages from debian, it has its own kernel team. It doesn't do a debug/release version of packages, it does stripped debug builds. All the packages that are in ubuntu are not in an alpha state, they're just current versions.
      It do not import everything from debian.

      They also build their own version based on svn sources.

      It's common to see ubuntu packages with svn in their version.

      All canonical supported packages are not only "imported" from debian.
      Sources are !
      That's why they do all this stages before getting a release.
      They are not only waiting for new packages to come to debian and then integrate them and hopefully I would say.

      All devs will tell you that during a testing stage there's no insurance that there's not some code here or there that will slow down the all thing.
      So you can't only base your judgement on beta/alpha's benchmarks.

      I'm not here to say that this test is useful or not. Just to attract attention on this point.

      Comment


      • #18
        Originally posted by hax0r View Post
        And why not benchmark alpha? It's not like the final release will run 3x faster, the numbers pretty much reflect actual performance that likely will not change.

        Who is exactly affected by ext4 bug? I'm using "data=writeback,noatime,nodiratime" options.
        YOU ARE AFFECTED!

        You have to add ",barrier=1" to your options.
        Otherwise there is a big performance penalty.
        Some have said it is of the order of 15 times slower than before...

        Not a thing to forget!

        But this more agressive setting will have a price. The EXT4 journaling may becomes unreliable... If you lose power at the wrong time, at reboot the system will apply the transactions not done from the journal but the journal will be corrupted ... This will hose your file system for sure. I'm going to move everything to reiser4 or something else soon if there is no solution.

        Do we need a journal for the journal ??? Ask the EXT4 guys they must be pulling out their hairs right now!!!

        Regards,

        BT

        Comment


        • #19
          I don't know how useful benchmarks are for the initial alpha versions of distributions. I would fully expect that speed of bootup is the least of the team's worries at this point. The team has put it in the spotlight yes, but bootup speed is not something that just happens when you throw everything together and once they have the right components then they can start worrying about speed.

          Comment


          • #20
            Originally posted by PerfMonk View Post
            Do we need a journal for the journal ??? Ask the EXT4 guys they must be pulling out their hairs right now!!!
            Or ask the rfs4 guys, they must be laughing at the ext4 ones right now. Recovering a filesystem correctly from a software crash is a solved problem. Giving a useful level of data integrity without running slower than molasses has been done too - ext3 does it. I don't know how these guys managed to screw a formerly good filesystem up so badly.

            Comment


            • #21
              9.10 is stable??

              "the current stable release, Ubuntu 9.10."
              I would hardly call Karmic Koala stable!!
              Half the time my sound pops and crackles the other half of the time I get no sound.
              I was trying to use the Gimp the other day and my system totally crashed. Multiple times.
              The manual eject button on my DVD drive no longer works.
              Transmission, Sauerbraten and others hang on exit and I have to kill X server.
              I think it is high time Ubuntu ditches the 6 month release cycle and does some heavy testing between releases!

              Comment


              • #22
                Originally posted by PerfMonk View Post
                YOU ARE AFFECTED!

                Originally posted by PerfMonk View Post
                You have to add ",barrier=1" to your options.
                Otherwise there is a big performance penalty.
                Thanks! I should start reading lkml.org . Maybe it's time for reiser4, but it looks like it's not going to get merged into 2.6.33. Btw, welcome to phoronix.org

                Comment


                • #23
                  Originally posted by Ant P. View Post
                  Giving a useful level of data integrity without running slower than molasses has been done too - ext3 does it.
                  Not exactly: as I understand it one of the biggest performance problems with Linux currently is that an fsync() on ext3 to flush a two-byte file to disk will flush every single byte of cached write data for every single file that's being written to out to disk before it complete... so that two-byte flush might result in 100+MB of writes to multiple different locations on the disk.

                  Ultimately, getting good performance and reliable data storage at the same time is hard.

                  Comment


                  • #24
                    Originally posted by hax0r View Post
                    Who is exactly affected by ext4 bug? I'm using "data=writeback,noatime,nodiratime" options.
                    It's not a bug.
                    Last edited by kraftman; 12-10-2009, 02:05 AM.

                    Comment


                    • #25
                      Originally posted by cliff View Post
                      "the current stable release, Ubuntu 9.10."
                      I would hardly call Karmic Koala stable!!
                      Half the time my sound pops and crackles the other half of the time I get no sound.
                      I was trying to use the Gimp the other day and my system totally crashed. Multiple times.
                      The manual eject button on my DVD drive no longer works.
                      Transmission, Sauerbraten and others hang on exit and I have to kill X server.
                      I think it is high time Ubuntu ditches the 6 month release cycle and does some heavy testing between releases!
                      Better check if your hardware is stable. This manual eject button works in Kubuntu and afaik didn't work in Ubuntu since release (it's probably DE problem or something - in KDE there's a service running which let you to eject your DVD manually). Btw. Kubuntu is rock stable here, so like I said it can be something with your hardware.

                      Comment


                      • #26
                        Just some precisions,

                        I reported the journal as being corrupted : I was wrong.
                        This is the FileSystem that can get corrupted even if the journal is not. Let me explain... The new drives will try to optimise drive access by ordering writes physically close together. If you have let say 3 writes to do, then the number 3 can be written before the number 2 because it is physically closer to actual position of the head. This does not play well with a journaling filesystem. You have to write something first in the journal and then wite to the disk. But if for optimisation the write to a bloc of a file is written before the actual write of the journal and if the system lose power in between those 2 writes, your filesystem can become inconsistent even if EXT4 has a journal and a checksum to validate the journal content. This is your filesystem that become incorrect, your journal has never seen this change you did...

                        I'm sure that Mr. T'so and all the EXT4 guys are working hard to correct this ASAP (see http://git.kernel.org/?p=linux/kerne...60256d1d726fdc)

                        Regards,

                        BT

                        Comment


                        • #27
                          Ubuntu 10.04 Is Off To A Poor Performance Start

                          The title is alarmist!!

                          EXT3 or EXT2 or XFS or BTRFS or ... + nvidia or intel gfx drivers = Good performance

                          The exception, ext4 (with barriers) and ati drivers (open source).

                          Comment


                          • #28
                            weird HW used

                            First of all, thank you very much for your efford in linux testing and benchmarking, I finally found great server that cares about linux benchmarks.
                            Just one question: why you run this test on overclocked CPU and very small and very slow HDD? I dont say that using some "normal" and fast HDD like WD WD6400AAK would produce totally different results, but combination of very fast overclocked CPU and very slow HDD is a little bit weird in my opinion.

                            Comment

                            Working...
                            X