[lkml]   [2001]   [Nov]   [6]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
    SubjectRe: [Ext2-devel] disk throughput

    On Mon, 5 Nov 2001, Alexander Viro wrote:
    > Oh, come on. (a) is obvious, but obviously not enough ;-)

    I agree, but I think it's a fairly good starting point to build up some

    Also note that traditonal UNIX implementations will have a hard time doing
    anything more fine-grained than "is the parent the root directory or not".
    Information about grandparents etc has been lost long before we get to
    "ialloc()", so I bet you'll find experimental data only on that one-level

    But it is probably not a binary decision in real life.

    In real life, you're ok switching cylinder groups for /home vs /usr, but
    you're also ok switching groups for /home/torvalds vs /home/viro. And it's
    still reasonably likely that it makes sense to switch groups in
    /home/torvalds between 'src' and 'public_html'.

    But the deeper down in the directory hierarchy you get, the less likely it
    is that it makes sense.

    And yes, it's just a heuristic. But it's one that where thanks to the
    dcache we could reasonably easily do more than just a black-and-white
    "root vs non-root" thing.

    > Yes, but block reservation also makes sense (otherwise we can end up
    > putting a directory into parent's CG only to have all children
    > going there _and_ getting far from their data). Which might be the
    > problem with original code, BTW.

    Yes, agreed. Disk space allocations should definitely be part of the
    heuristics, and that's obviously both inode and blocks.

    I do believe that there are probably more "high-level" heuristics that can
    be useful, though. Looking at man common big trees, the ownership issue is
    one obvious clue. Sadly the trees obviously end up being _created_ without
    owner information, and the chown/chgrp happens later, but it might still
    be useable for some clues.


    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to
    More majordomo info at
    Please read the FAQ at

     \ /
      Last update: 2005-03-22 13:12    [W:0.024 / U:31.844 seconds]
    ©2003-2016 Jasper Spaans. hosted at Digital OceanAdvertise on this site