[lkml]   [2008]   [Feb]   [8]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
    SubjectRe: IO queuing and complete affinity with threads (was Re: [PATCH 0/8] IO queuing and complete affinity)
    On Fri, Feb 08, 2008 at 08:47:47AM +0100, Jens Axboe wrote:
    > On Fri, Feb 08 2008, Nick Piggin wrote:
    > > On Thu, Feb 07, 2008 at 07:25:45PM +0100, Jens Axboe wrote:
    > > > Hi,
    > > >
    > > > Here's a variant using kernel threads only, the nasty arch bits are then
    > > > not needed. Works for me, no performance testing (that's a hint for Alan
    > > > to try and queue up some testing for this variant as well :-)
    > >
    > > Well this stuff looks pretty nice (although I'm not sure whether the
    > > softirq->thread changes are a good idea for performance, I guess we'll
    > > see).
    > Yeah, that is indeed an open question and why I have two seperate
    > patches for now (io-cpu-affinity branch and io-cpu-affinity-kthread
    > branch). As Ingo mentioned, this is how softirqs are handled in the -rt
    > branch already.

    True, although there are some IO workloads where -rt falls behind
    mainline. May not be purely due to irq threads though, of course.

    > > You still don't have the option that the Intel patch gave, that is,
    > > to submit on the completer. I guess that you could do it somewhat
    > > generically by having a cpuid in the request queue, and update that
    > > with the completing cpu.
    > Not sure what you mean, if setting queue_affinity doesn't accomplish it.
    > If you know the completing CPU to begin with, surely you can just set
    > the queuing affinity appropriately?

    And if you don't?

    > > At least they reported it to be the most efficient scheme in their
    > > testing, and Dave thought that migrating completions out to submitters
    > > might be a bottleneck in some cases.
    > More so than migrating submitters to completers? The advantage of only
    > movign submitters is that you get rid of the completion locking. Apart
    > from that, the cost should be the same, especially for the thread based
    > solution.

    Not specifically for the block layer, but higher layers like xfs.

     \ /
      Last update: 2008-02-08 08:55    [W:0.032 / U:4.492 seconds]
    ©2003-2017 Jasper Spaans. hosted at Digital OceanAdvertise on this site