Announcement

Collapse
No announcement yet.

The Linux Desktop Responsiveness Patches Are Feeling Good

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

  • #31
    I'm running the latest zen kernel, and the kill_machine program still does manage to kill my machine if there's any swap... If there's none (swapoff -a) it just hangs it for a few seconds before it's killed for running out of ram. It seems to me like swap isn't counted by the io scheduler as process-specific. Moreover swapping seems to have the highest priority for hdd access. Maybe it doesn't even go through the io scheduler at all; it seems like the scheduler only has what's left from swapping to work with. I'm wondering why the hell that is. Seems like a very, um, silly thing for the kernel to do.

    Comment


    • #32
      yeah, still some jerkiness - but only in the special case of using zfs-fuse on top of a device-mapper (LUKS) device (with gzip-encryption enabled, sha256-checksumming) and scrubbing running (on around 750 GiB of data)

      even 2.6.35 with CONFIG_CRYPTO_PCRYPT and tweaked /etc/zfs/zfsrc doesn't help

      try to play internet-radio via amarok and you will enjoy stuttering sound at it's best (seems like this triggers some kind of problem with amarok)

      since this is the only app with this issue - flash-based players, gstreamer-& mono-based banshee, etc. work fine

      Comment


      • #33
        Originally posted by kernelOfTruth View Post
        yeah, still some jerkiness - but only in the special case of using zfs-fuse on top of a device-mapper (LUKS) device (with gzip-encryption enabled, sha256-checksumming) and scrubbing running (on around 750 GiB of data)

        even 2.6.35 with CONFIG_CRYPTO_PCRYPT and tweaked /etc/zfs/zfsrc doesn't help

        try to play internet-radio via amarok and you will enjoy stuttering sound at it's best (seems like this triggers some kind of problem with amarok)

        since this is the only app with this issue - flash-based players, gstreamer-& mono-based banshee, etc. work fine
        forget that

        it was a phonon / KDE4 related problem:

        switching from the gstreamer to the xine-backend "fixed" it

        so nearly no stalls anymore

        and I'm sure 2.6.36 will be even better (lots of improvements already got in according to the postings on lkml)

        Comment


        • #34
          Originally posted by kernelOfTruth View Post
          forget that

          it was a phonon / KDE4 related problem:

          switching from the gstreamer to the xine-backend "fixed" it

          so nearly no stalls anymore

          and I'm sure 2.6.36 will be even better (lots of improvements already got in according to the postings on lkml)
          I got phonon fixed by removing pulseaudio from my system. phonon-gstreamer seems to work fine with plain alsa, and phonon-xine causes some problems with Amarok. Plus, I don't think xine is being actively developed by anyone anymore...

          Comment


          • #35
            Originally posted by loonyphoenix View Post
            I got phonon fixed by removing pulseaudio from my system. phonon-gstreamer seems to work fine with plain alsa, and phonon-xine causes some problems with Amarok. Plus, I don't think xine is being actively developed by anyone anymore...
            how would you know ?

            anyways: pulseaudio is working fine now for me and I'm enjoying the additional comfort (since the kernel largely won't get into lagging anymore)

            yeah, it's really a shame that xine isn't (?) developed anymore (I'm referring to the xine GUI since using it practically isn't worth it - often times it'll hardlock my box)

            xine-lib (xine ?) however is still developed: http://www.xine-project.org/home

            Latest news…

            2010-07-25: Release
            xine-lib 1.1.19
            A new version of xine-lib is available. This is mainly bug fixes, but there are one or two new features, notably support for the WebM container format.

            Please check the release notes for details & known problems.

            Download it from our download page.

            2010-03-06: Site news
            #xine-vdpau
            #xine-vdpau (freenode) is no more. Use #xine (alternatively, on OFTC) instead.

            It has been determined that, while it was the more active channel, a lot of its content was general xine-related chat, which is more on-topic in #xine, so we've decided that VDPAU no longer needs a separate channel. If you have a problem with this, send your comments to /dev/null ☺.

            2010-03-06: Release
            xine-lib 1.1.18.1
            A new version of xine-lib is available. This is a bug-fix release; there were a few ‘minor’ breakages in 1.1.18.

            Please check the release notes for details & known problems.

            Download it from our download page.

            Comment


            • #36
              Originally posted by kernelOfTruth View Post
              how would you know ?

              anyways: pulseaudio is working fine now for me and I'm enjoying the additional comfort (since the kernel largely won't get into lagging anymore)

              yeah, it's really a shame that xine isn't (?) developed anymore (I'm referring to the xine GUI since using it practically isn't worth it - often times it'll hardlock my box)

              xine-lib (xine ?) however is still developed: http://www.xine-project.org/home
              Sorry, my mistake then. I was sure I'd seen mention it had stopped active development.

              Comment


              • #37
                Originally posted by RealNC View Post
                I'm still looking for the full set of those patches for 2.6.35. Where are they?
                Me as well.

                Comment


                • #38
                  Originally posted by Ranguvar View Post
                  Me as well.
                  are you guys referring to lkml.org or lwn.net ?

                  if not:

                  patch 1&2:
                  [PATCH mmotm] vmscan: raise the bar to PAGEOUT_IO_SYNC stalls

                  fix:
                  vmscan: fix inconsistency reported by Kail and confirmed by Wu Fengguang

                  that's all you need for 2.6.35 ! (it already includes the other stuff from patch 5-7)

                  the 2.6.35 branch is the way to go - only the first 2 patches are needed for that because it already includes the additional ones

                  for 2.6.34 you need a lot more but it's also included with zen-stable

                  Comment


                  • #39
                    besides that it really DOES make a difference:

                    I'm currently re-testing usage of reiser4 and there are almost no stalls / hickups during webradio streaming when the filesystems needs to heavily flush things to disk (even with device-mapper & LUKS) - which definitely was a problem in the past

                    during heavy swapping of 1-2 GiBs (e.g. compiling chromium in ram) it's still somewhat responsive but sound is broken for several seconds-minutes (pulseaudio obviously significantly makes it worse) and the GUI also doesn't react for several (dozen) seconds

                    so as a summary:
                    the management of swapping out stuff needs some more improvements in desktop land; flushing and handling of the filesystems in general has greatly improved

                    Comment


                    • #40
                      One question folks:

                      I got two sets of patches. One has
                      Code:
                      +		/* Check if we should synchronously wait for writeback */
                      +		if (should_reclaim_stall(nr_taken, nr_reclaimed, priority, sc)) {
                      and the other has:
                      Code:
                      +		/* Check if we should syncronously wait for writeback */
                      +		if (should_reclaim_stall(nr_taken, nr_freed, priority, sc)) {
                      Which one of them is right for 2.6.35?

                      Comment

                      Working...
                      X