[lkml]   [1997]   [Nov]   [24]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
    SubjectRe: More on 2.1.65 slowdown (disc thrashing)
    On Fri, 21 Nov 1997, Janos Farkas wrote:

    > Just recently, fsync has been changed to unconditionally call sync(),
    > because it has been found that fsync on very large files can take
    > enormous times (even hours!). This is because it wanted to sync all
    > the indirect blocks in addition to the changed data blocks, so it
    > could be unbearably slow.

    Unacceptable when a device sync is faster :)

    > Now, it seems making this to sync() always
    > is not too wise either. Possibly the best thing would be to do
    > this only on substantially large files (>64M or so). Maybe making
    > fdatasync() work in the same shot would be nice too, before the VFS
    > interface will be frozen...

    Well, we could just fsync all files with:
    - only direct and single indirect blocks
    - files that only have <= xx dirty blocks in-core

    And files with double / triple indirect blocks can be
    dev-synced (maybe even an 'asynchronous sync'... :-)


    Send Linux memory-management wishes to me: I'm currently looking
    for something to hack...

     \ /
      Last update: 2005-03-22 13:40    [W:0.031 / U:37.680 seconds]
    ©2003-2016 Jasper Spaans. hosted at Digital OceanAdvertise on this site