lkml.org 
[lkml]   [2009]   [Dec]   [10]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    Date
    From
    Subject[patch 0/2] sched: Change nohz ilb logic from pull to push model
    This is a followup to the RFC here:
    http://lkml.indiana.edu/hypermail/linux/kernel/0906.2/01163.html

    We have few cleanups since that RFC and also have some data
    showing the impact of this change.

    Description:
    Existing nohz idle load balance logic uses the pull model, with one
    idle load balancer CPU nominated on any partially idle system and that
    balancer CPU not going into nohz mode. With the periodic tick, the
    balancer does the idle balancing on behalf of all the CPUs in nohz mode.

    This is not very optimal and has few issues:
    * the balancer will continue to have periodic ticks and wakeup
    frequently (HZ rate), even though it may not have any rebalancing to do on
    behalf of any of the idle CPUs.
    * On x86 and CPUs that have APIC timer stoppage on idle CPUs, this periodic
    wakeup can result in an additional interrupt on a CPU doing the timer
    broadcast.
    * The balancer may end up spending a lot of time doing the balancing on
    behalf of nohz CPUs, especially with increasing number of sockets and
    cores in the platform.

    The alternative is to have a push model, where all idle CPUs can enter nohz
    mode and any busy CPU kicks one of the idle CPUs to take care of idle
    balancing on behalf of a group of idle CPUs.

    Following patches switches idle load balancer to this push approach.

    Data:
    1) Running a bunzip2 of a big file (which happened to be kernel tar ball),
    on a netbook with HZ=1000.

    Before the change
    57.44user 12.36system 1:12.17elapsed

    After the change
    47.89user 10.31system 0:59.99elapsed

    That is ~10 seconds (17%) savings in user time for this task. This is
    coming from the idle SMT sibling thread being woken up 1000 times a second
    and doing unnecessary idle load balancing, resulting in
    slowdown of the thread running the load.

    2) Running bzip2 of a big file (which happened to be kernel tar ball),
    on a dual socket server with HZ=1000

    No change in performance, but there is a noticable (1% - 1.5% range)
    reduction in energy consumption. This is due to idle load balancer
    that un necessarily gets woken up on second socket with the earlier pull
    model. With new push model, second socket will not get woken up often
    and can get into low power idle state.

    3) We also measured SpecJBB workload with varying number of warehouses
    and did not see any noticable change in performance with this patch.

    Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
    Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>

    --



    \
     
     \ /
      Last update: 2009-12-11 02:43    [W:3.259 / U:0.056 seconds]
    ©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site