Re: [PATCH -v2] mm: do batched scans for mem_cgroup

[Wu Fengguang <fengguang.wu@intel.com>]

On Thu, Aug 20, 2009 at 01:16:56PM +0800, Balbir Singh wrote:
> * Wu Fengguang <fengguang.wu@intel.com> [2009-08-20 12:05:33]:
> 
> > On Thu, Aug 20, 2009 at 11:13:47AM +0800, KAMEZAWA Hiroyuki wrote:
> > > On Thu, 20 Aug 2009 10:49:29 +0800
> > > Wu Fengguang <fengguang.wu@intel.com> wrote:
> > > 
> > > > For mem_cgroup, shrink_zone() may call shrink_list() with nr_to_scan=1,
> > > > in which case shrink_list() _still_ calls isolate_pages() with the much
> > > > larger SWAP_CLUSTER_MAX.  It effectively scales up the inactive list
> > > > scan rate by up to 32 times.
> > > > 
> > > > For example, with 16k inactive pages and DEF_PRIORITY=12, (16k >> 12)=4.
> > > > So when shrink_zone() expects to scan 4 pages in the active/inactive
> > > > list, it will be scanned SWAP_CLUSTER_MAX=32 pages in effect.
> > > > 
> > > > The accesses to nr_saved_scan are not lock protected and so not 100%
> > > > accurate, however we can tolerate small errors and the resulted small
> > > > imbalanced scan rates between zones.
> > > > 
> > > > This batching won't blur up the cgroup limits, since it is driven by
> > > > "pages reclaimed" rather than "pages scanned". When shrink_zone()
> > > > decides to cancel (and save) one smallish scan, it may well be called
> > > > again to accumulate up nr_saved_scan.
> > > > 
> > > > It could possibly be a problem for some tiny mem_cgroup (which may be
> > > > _full_ scanned too much times in order to accumulate up nr_saved_scan).
> > > > 
> > > > CC: Rik van Riel <riel@redhat.com>
> > > > CC: Minchan Kim <minchan.kim@gmail.com>
> > > > CC: Balbir Singh <balbir@linux.vnet.ibm.com>
> > > > CC: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
> > > > CC: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
> > > > Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
> > > > ---
> > > 
> > > Hmm, how about this ? 
> > > ==
> > > Now, nr_saved_scan is tied to zone's LRU.
> > > But, considering how vmscan works, it should be tied to reclaim_stat.
> > > 
> > > By this, memcg can make use of nr_saved_scan information seamlessly.
> > 
> > Good idea, full patch updated with your signed-off-by :)
> > 
> > Thanks,
> > Fengguang
> > ---
> > mm: do batched scans for mem_cgroup
> > 
> > For mem_cgroup, shrink_zone() may call shrink_list() with nr_to_scan=1,
> > in which case shrink_list() _still_ calls isolate_pages() with the much
> > larger SWAP_CLUSTER_MAX.  It effectively scales up the inactive list
> > scan rate by up to 32 times.
> > 
> > For example, with 16k inactive pages and DEF_PRIORITY=12, (16k >> 12)=4.
> > So when shrink_zone() expects to scan 4 pages in the active/inactive
> > list, it will be scanned SWAP_CLUSTER_MAX=32 pages in effect.
> > 
> > The accesses to nr_saved_scan are not lock protected and so not 100%
> > accurate, however we can tolerate small errors and the resulted small
> > imbalanced scan rates between zones.
> > 
> > This batching won't blur up the cgroup limits, since it is driven by
> > "pages reclaimed" rather than "pages scanned". When shrink_zone()
> > decides to cancel (and save) one smallish scan, it may well be called
> > again to accumulate up nr_saved_scan.
> > 
> > It could possibly be a problem for some tiny mem_cgroup (which may be
> > _full_ scanned too much times in order to accumulate up nr_saved_scan).
> >
> 
> Looks good to me, how did you test it?

I observed the shrink_inactive_list() calls with this patch:

        @@ -1043,6 +1043,13 @@ static unsigned long shrink_inactive_lis
                struct zone_reclaim_stat *reclaim_stat = get_reclaim_stat(zone, sc);
                int lumpy_reclaim = 0;

        +       if (!scanning_global_lru(sc))
        +               printk("shrink inactive %s count=%lu scan=%lu\n",
        +                      file ? "file" : "anon",
        +                      mem_cgroup_zone_nr_pages(sc->mem_cgroup, zone,
        +                                               LRU_INACTIVE_ANON + 2 * !!file),
        +                      max_scan);

and these commands:

        mkdir /cgroup/0
        echo 100M > /cgroup/0/memory.limit_in_bytes
        echo $$ > /cgroup/0/tasks
        cp /tmp/10G /dev/null

before patch:

        [ 3682.646008] shrink inactive file count=25535 scan=6
        [ 3682.661548] shrink inactive file count=25535 scan=6
        [ 3682.666933] shrink inactive file count=25535 scan=6
        [ 3682.682865] shrink inactive file count=25535 scan=6
        [ 3682.688572] shrink inactive file count=25535 scan=6
        [ 3682.703908] shrink inactive file count=25535 scan=6
        [ 3682.709431] shrink inactive file count=25535 scan=6

after patch:

        [  223.146544] shrink inactive file count=25531 scan=32
        [  223.152060] shrink inactive file count=25507 scan=10
        [  223.167503] shrink inactive file count=25531 scan=32
        [  223.173426] shrink inactive file count=25507 scan=10
        [  223.188764] shrink inactive file count=25531 scan=32
        [  223.194270] shrink inactive file count=25507 scan=10
        [  223.209885] shrink inactive file count=25531 scan=32
        [  223.215388] shrink inactive file count=25507 scan=10

Before patch, the inactive list is over scanned by 30/6=5 times;
After patch, it is over scanned by 64/42=1.5 times. It's much better,
and can be further improved if necessary.

> Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>

Thanks!

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>