NFS/sound problems appear to be due to huge latency with kswapd

From: Russell King (rmk@arm.linux.org.uk)
Date: Sat Mar 25 2000 - 17:02:05 EST

  • Next message: delete_list@YAHOO.COM: "50 Megs of web space for $9.95 a month"

    Hi,

    Over the past week, I've posted various messages to Linux kernel about
    various NFS and sound problems on 2.3.99, some of them have turned out
    to be bugs in the ARM code, but there is one in particular which has
    been difficult to find.

    This problem is where I play mp3s on a machine, and the mp3s are stored
    remotely.

    By adding various debugs and so forth to the kernel, I think I've found
    the culprit: kswapd.

    kswapd appears to take a long time to complete its task regularly (200ms+
    on occasion). Here is the messages that indicate this:

    0000e943: RPC: 2426 removed from queue c103a048 "xprt_pending"
    0000e943: RPC: 2426 added to queue c01ad5d4 "schedq"
    0000e943: need_resched = 1, comm = mxaudio
    0000e943: RPC: __rpc_wake_up done
    kswapd: awake at 59715
    0000e943: RPC: 2427 __rpc_wake_up (now 59715 inh 0)
    0000e943: RPC: 2427 __rpc_wake_up (now 59715 inh 0)
    0000e943: RPC: 2427 removed from queue c103a048 "xprt_pending"
    0000e943: RPC: 2427 added to queue c01ad5d4 "schedq"
    0000e943: need_resched = 0, comm = kswapd
    0000e943: RPC: __rpc_wake_up done
    kswapd: sleep at 59744
    0000e960: RPC: switch to rpciod
    0000e960: RPC: rpc_schedule enter
    ...
    kswapd: awake at 65355
    kswapd: sleep at 65394
    ...
    kswapd: awake at 79811
    kswapd: sleep at 79812
    ...
    kswapd: awake at 82853
    kswapd: sleep at 82892
    ...
    kswapd: awake at 102281
    kswapd: sleep at 102320

    Sometimes it takes around 10ms to do its job, other times it can be as high
    as 400ms. Obviously this is not good for interactive performance, let alone
    using anything like sound under Linux.

    I'm not sure if this is an ARM specific problem, but I wouldn't have thought
    so. It may be related to the problem where the cached data takes over all
    the memory, but I'm not sure.

    /proc/meminfo indicates:

            total: used: free: shared: buffers: cached:
    Mem: 30887936 30339072 548864 0 86016 20320256
    Swap: 66056192 0 66056192
    ...
       _____
      |_____| ------------------------------------------------- ---+---+-
      | | Russell King rmk@arm.linux.org.uk --- ---
      | | | | http://www.arm.linux.org.uk/~rmk/aboutme.html / / |
      | +-+-+ --- -+-
      / | THE developer of ARM Linux |+| /|\
     / | | | --- |
        +-+-+ ------------------------------------------------- /\\\ |

    -
    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to majordomo@vger.rutgers.edu
    Please read the FAQ at http://www.tux.org/lkml/

    This archive was generated by hypermail 2b29 : Sat Mar 25 2000 - 17:09:03 EST