Announcement

Collapse
No announcement yet.

Kolivas Pushes New Kernel Responsiveness Patches

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

  • #31
    if it is all about disk - why does a big copy make the mouse laggy? Hello? Mouse? Surely must be all the data that the mouse has to read from the disk all the time.
    Why is switching from X to vt or back goddamn slow when accessing a damaged disk at the same time? Why is waiting for ide timeouts making your nose bleed in anger?

    http://forums.gentoo.org/viewtopic-t...c-start-0.html
    http://forums.gentoo.org/viewtopic-t-793263.html

    those are 40 pages of suffering and trying stuff. Without much results.
    It is not even a new problem. This goes back to 2.2.10 (the kernel I started with) and never became better...

    Comment


    • #32
      Originally posted by energyman View Post
      if it is all about disk - why does a big copy make the mouse laggy? Hello? Mouse? Surely must be all the data that the mouse has to read from the disk all the time.
      Why is switching from X to vt or back goddamn slow when accessing a damaged disk at the same time? Why is waiting for ide timeouts making your nose bleed in anger?
      Because some people are complaining about CFS not about CFQ

      Comment


      • #33
        Originally posted by energyman View Post
        if it is all about disk - why does a big copy make the mouse laggy? .

        That's likely neither the process scheduler or the disk scheduler, but the VM subsystem.... it threw out the code pages for the mouse X code.. and now it needs to be read back from disk.

        Comment


        • #34
          that would explain an initial lag - but not the following stuttering. Also why should the VM discard code pages with tons of memory available?
          And seriously, discarding code of running programms sounds really really wrong.

          Comment


          • #35
            I don't think it's either a CPU scheduler nor a disk scheduler issue. If it were a problem with the disk scheduler, it would only affect other processes also accessing the disk, which plainly isn't the case. I also don't think it's the CPU scheduler, because I'd think in that case with a dual core CPU it could just put the disk stuff on one core and let the other stuff keep running on the other core. But that's not what happens. Even with two cores shit gets lagged up badly when doing disk-intensive stuff.

            I also don't think it's the VM subsystem... granted, the VM subsystem has its own pathologies (how long will it be before we can finally get it to never, ever favor file cache over application data?), but this happens even when there's plenty of physical memory free.

            As I said, I suspect it's something structural. But I'm not a kernel developer, so other than speculating "it's probably not X, because...", I can't really say.

            Comment


            • #36
              Originally posted by illissius View Post
              As I said, I suspect it's something structural. But I'm not a kernel developer, so other than speculating "it's probably not X, because...", I can't really say.

              if you can get timechart set up (part of the perf program), you can do

              perf timechart record sleep 10
              perf timechart

              the first command will do a 10 second recording of the activity in your system
              the second command will turn the recording into a svg image.

              with that, it's maybe possible to figure out what is going on, assuming the actual stuttering can be captured (if it's very reproducable, it should be)

              Comment


              • #37
                ok, I copied one 4.4gb iso from one md raid5 partition to another - /mnt/date to /var. system is on a SSD.
                At first it was ok. Then the mouse hanged twice will doing that perf command.

                It did so a few more times after those 10 seconds. Every couple of seconds the mouse hangs.

                I have to admit, it got better. Not too long ago it would have been horrible. - 8gb ram helped a lot.

                And now I have a huuuuge svg. I am uploading it right now at rapidshare, if someone is interessted.

                Comment


                • #38
                  Well been running 2.6.33-ck1 for the last 6h.
                  Done some gentoo update with no noticable change

                  QL however ran smoother!!!
                  No noticable difference in HoN

                  so seems good will be sticking with it

                  Comment


                  • #39
                    Originally posted by Naib View Post
                    Well been running 2.6.33-ck1 for the last 6h.
                    Done some gentoo update with no noticable change

                    QL however ran smoother!!!
                    No noticable difference in HoN

                    so seems good will be sticking with it
                    Zen kernel have the ck patchset included. (Btw, I'm using zen kernel since year ago, works great.)

                    Comment


                    • #40
                      Originally posted by RealNC View Post
                      It will probably never be merged. The majority of the kernel devs agreed that there won't be more than one CPU scheduler in the kernel.
                      If one is better for desktop use, and the other for server use, maybe both should be? One that did best under both situations would be ideal though of course.

                      Comment


                      • #41
                        Originally posted by Yfrwlf View Post
                        If one is better for desktop use, and the other for server use, maybe both should be? One that did best under both situations would be ideal though of course.
                        Who will judge which is better for desktop? For me it's definitely CFS.

                        Comment


                        • #42
                          Originally posted by kraftman View Post
                          Who will judge which is better for desktop?
                          How about Phoronix Test Suite?

                          Comment


                          • #43
                            Originally posted by arjan_intel View Post
                            if you can get timechart set up (part of the perf program), you can do

                            perf timechart record sleep 10
                            perf timechart

                            the first command will do a 10 second recording of the activity in your system
                            the second command will turn the recording into a svg image.

                            with that, it's maybe possible to figure out what is going on, assuming the actual stuttering can be captured (if it's very reproducable, it should be)
                            I thought this problem gone when I bought a new hard drive, but it's still present. I'm using 2.6.33 kernel from Ubuntu PPA. I caught some characteristic which doesn't look good (a lot of red on Plasma-desktop ;>). I will send you a link to svg file or maybe I should report it somewhere else? This is a screen shot if someone's interested:

                            http://img651.imageshack.us/img651/8334/latency.png

                            @Pejakm

                            How about Phoronix Test Suite?
                            If only some good tests will be used Last time BFS was faster at Phoronix, but at least one of the tests was meaningless - apache benchmark. If some problems with BFS are already solved and if someone will proof it's better on desktop then it will be nice to have them both in the mainline.

                            Comment


                            • #44
                              Originally posted by droidhacker View Post
                              Seems to me that more often than not, his work is unusably buggy and makes things explode in MS like frequency.
                              [citation needed]

                              Comment


                              • #45
                                Originally posted by pejakm View Post
                                How about Phoronix Test Suite?
                                Measuring Latency is not easy, benchmarks don't help here.

                                Comment

                                Working...
                                X