Announcement

Collapse
No announcement yet.

A New BFS "Smoking" Scheduler For Linux 3.3

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

  • #46
    Originally posted by evolution View Post
    Well, I'm going to give my personal experience of BFS:

    For gaming/compiling and video playback (in a 9 year old P4C), BFS is slightly better then CFS. For instance, if I'm loading a web page and watching a HD-video on mplayer-vaapi, while the browser loads the page, the audio of my video stutters a lot with CFS, whereas with BFS, audio stutters much less.
    How did BFS compare to CFS + nice?
    How did BFS compare to CFS + a decreased sched_granularity_ns

    I understand that setting the niceness of a browser-flash-plugin may be...... problematic (there's a better word than problematic, but it escapes me).

    F

    Comment


    • #47
      Originally posted by RealNC View Post
      Yo, peace mah man. Me an' mah homies are chillin' an' readin' thah forums. Yo, check it out man, people are writin' "peace" cuz it makes 'em look cool.
      You made my day twice on this thread

      Anyway, I would enjoy seeing some benchmarks about the BFQ disk scheduler. I've always wondered how it performs comparing to the default CFQ on recent kernels.

      Comment


      • #48
        Originally posted by russofris View Post
        How did BFS compare to CFS + nice?
        How did BFS compare to CFS + a decreased sched_granularity_ns

        I understand that setting the niceness of a browser-flash-plugin may be...... problematic (there's a better word than problematic, but it escapes me).

        F
        Well, the results I'm mentioning here are with kernel "defaults" (I didn't change any of those paramenters in "/proc/sys/", but I can test them...), the kernel compilation was done with "time make -j2" (P4 with HT) and time "make -j2 install".

        I was talking about doing playback of a full-HD video in mplayer-vaapi with 5.1 software mixing (I've a HD4650AGP that supports it) and, at the same time, doing some basic browser navigation with Opera+Flash...

        Cheers

        Comment


        • #49
          Originally posted by Paradox Uncreated View Post
          If the latency is 0.33ms, max time for short tasks = 0.33ms, and it is displayed the next frame. You cannot feel the difference here.
          But we can measure 16 milliseconds of time. Which is roughly the time it takes between frames on a "normal" display. Well, 8 is starting to get more common as the 3D rage pushes up the monitors framerates.
          A 0.33 ms advantage per task, perhaps 40-50 times a second, that will after some time produce really large numbers. And lets for arguments sake say we play some game, perhaps a emulator, where after some time the difference in troughput for just that application will be felt measurable. Uneven audio playback, cutscenes takes longer to finish, loading takes longer.
          Even if what we have at hand can not be measured by us directly, we do feel it. Just like that we can perceive the difference of 58-62 uneven framerate vs 60 stable frames.
          The reason BFS is better than CFS is because we do not care about all those background applications that is running when we are doing something, what we care about is not them getting troughput, we care about the pointer and the application window getting updates and redrawed exactly as many times as the desktop refreshes each second. Even post 12 points at this important aspect.

          Comment

          Working...
          X