lkml.org 
[lkml]   [2011]   [May]   [3]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    Date
    SubjectRe: [RFC][PATCH] mm: cut down __GFP_NORETRY page allocation failures
    From
    On Thu, Apr 28, 2011 at 9:36 PM, Wu Fengguang <fengguang.wu@intel.com> wrote:
    > Concurrent page allocations are suffering from high failure rates.
    >
    > On a 8p, 3GB ram test box, when reading 1000 sparse files of size 1GB,
    > the page allocation failures are
    >
    > nr_alloc_fail 733       # interleaved reads by 1 single task
    > nr_alloc_fail 11799     # concurrent reads by 1000 tasks
    >
    > The concurrent read test script is:
    >
    >        for i in `seq 1000`
    >        do
    >                truncate -s 1G /fs/sparse-$i
    >                dd if=/fs/sparse-$i of=/dev/null &
    >        done
    >

    With Core2 Duo, 3G ram, No swap partition I can not produce the alloc fail

    > In order for get_page_from_freelist() to get free page,
    >
    > (1) try_to_free_pages() should use much higher .nr_to_reclaim than the
    >    current SWAP_CLUSTER_MAX=32, in order to draw the zone out of the
    >    possible low watermark state as well as fill the pcp with enough free
    >    pages to overflow its high watermark.
    >
    > (2) the get_page_from_freelist() _after_ direct reclaim should use lower
    >    watermark than its normal invocations, so that it can reasonably
    >    "reserve" some free pages for itself and prevent other concurrent
    >    page allocators stealing all its reclaimed pages.
    >
    > Some notes:
    >
    > - commit 9ee493ce ("mm: page allocator: drain per-cpu lists after direct
    >  reclaim allocation fails") has the same target, however is obviously
    >  costly and less effective. It seems more clean to just remove the
    >  retry and drain code than to retain it.
    >
    > - it's a bit hacky to reclaim more than requested pages inside
    >  do_try_to_free_page(), and it won't help cgroup for now
    >
    > - it only aims to reduce failures when there are plenty of reclaimable
    >  pages, so it stops the opportunistic reclaim when scanned 2 times pages
    >
    > Test results:
    >
    > - the failure rate is pretty sensible to the page reclaim size,
    >  from 282 (WMARK_HIGH) to 704 (WMARK_MIN) to 10496 (SWAP_CLUSTER_MAX)
    >
    > - the IPIs are reduced by over 100 times
    >
    > base kernel: vanilla 2.6.39-rc3 + __GFP_NORETRY readahead page allocation patch
    > -------------------------------------------------------------------------------
    > nr_alloc_fail 10496
    > allocstall 1576602
    >
    > slabs_scanned 21632
    > kswapd_steal 4393382
    > kswapd_inodesteal 124
    > kswapd_low_wmark_hit_quickly 885
    > kswapd_high_wmark_hit_quickly 2321
    > kswapd_skip_congestion_wait 0
    > pageoutrun 29426
    >
    > CAL:     220449     220246     220372     220558     220251     219740     220043     219968   Function call interrupts
    >
    > LOC:     536274     532529     531734     536801     536510     533676     534853     532038   Local timer interrupts
    > RES:       3032       2128       1792       1765       2184       1703       1754       1865   Rescheduling interrupts
    > TLB:        189         15         13         17         64        294         97         63   TLB shootdowns

    Could you tell how to get above info?

    >
    > patched (WMARK_MIN)
    > -------------------
    > nr_alloc_fail 704
    > allocstall 105551
    >
    > slabs_scanned 33280
    > kswapd_steal 4525537
    > kswapd_inodesteal 187
    > kswapd_low_wmark_hit_quickly 4980
    > kswapd_high_wmark_hit_quickly 2573
    > kswapd_skip_congestion_wait 0
    > pageoutrun 35429
    >
    > CAL:         93        286        396        754        272        297        275        281   Function call interrupts
    >
    > LOC:     520550     517751     517043     522016     520302     518479     519329     517179   Local timer interrupts
    > RES:       2131       1371       1376       1269       1390       1181       1409       1280   Rescheduling interrupts
    > TLB:        280         26         27         30         65        305        134         75   TLB shootdowns
    >
    > patched (WMARK_HIGH)
    > --------------------
    > nr_alloc_fail 282
    > allocstall 53860
    >
    > slabs_scanned 23936
    > kswapd_steal 4561178
    > kswapd_inodesteal 0
    > kswapd_low_wmark_hit_quickly 2760
    > kswapd_high_wmark_hit_quickly 1748
    > kswapd_skip_congestion_wait 0
    > pageoutrun 32639
    >
    > CAL:         93        463        410        540        298        282        272        306   Function call interrupts
    >
    > LOC:     513956     510749     509890     514897     514300     512392     512825     510574   Local timer interrupts
    > RES:       1174       2081       1411       1320       1742       2683       1380       1230   Rescheduling interrupts
    > TLB:        274         21         19         22         57        317        131         61   TLB shootdowns
    >
    > this patch (WMARK_HIGH, limited scan)
    > -------------------------------------
    > nr_alloc_fail 276
    > allocstall 54034
    >
    > slabs_scanned 24320
    > kswapd_steal 4507482
    > kswapd_inodesteal 262
    > kswapd_low_wmark_hit_quickly 2638
    > kswapd_high_wmark_hit_quickly 1710
    > kswapd_skip_congestion_wait 0
    > pageoutrun 32182
    >
    > CAL:         69        443        421        567        273        279        269        334   Function call interrupts
    >
    > LOC:     514736     511698     510993     514069     514185     512986     513838     511229   Local timer interrupts
    > RES:       2153       1556       1126       1351       3047       1554       1131       1560   Rescheduling interrupts
    > TLB:        209         26         20         15         71        315        117         71   TLB shootdowns
    >
    > CC: Mel Gorman <mel@linux.vnet.ibm.com>
    > Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
    > ---
    >  mm/page_alloc.c |   17 +++--------------
    >  mm/vmscan.c     |    6 ++++++
    >  2 files changed, 9 insertions(+), 14 deletions(-)
    > --- linux-next.orig/mm/vmscan.c 2011-04-28 21:16:16.000000000 +0800
    > +++ linux-next/mm/vmscan.c      2011-04-28 21:28:57.000000000 +0800
    > @@ -1978,6 +1978,8 @@ static void shrink_zones(int priority, s
    >                                continue;
    >                        if (zone->all_unreclaimable && priority != DEF_PRIORITY)
    >                                continue;       /* Let kswapd poll it */
    > +                       sc->nr_to_reclaim = max(sc->nr_to_reclaim,
    > +                                               zone->watermark[WMARK_HIGH]);
    >                }
    >
    >                shrink_zone(priority, zone, sc);
    > @@ -2034,6 +2036,7 @@ static unsigned long do_try_to_free_page
    >        struct zoneref *z;
    >        struct zone *zone;
    >        unsigned long writeback_threshold;
    > +       unsigned long min_reclaim = sc->nr_to_reclaim;
    >
    >        get_mems_allowed();
    >        delayacct_freepages_start();
    > @@ -2067,6 +2070,9 @@ static unsigned long do_try_to_free_page
    >                        }
    >                }
    >                total_scanned += sc->nr_scanned;
    > +               if (sc->nr_reclaimed >= min_reclaim &&
    > +                   total_scanned > 2 * sc->nr_to_reclaim)
    > +                       goto out;
    >                if (sc->nr_reclaimed >= sc->nr_to_reclaim)
    >                        goto out;
    >
    > --- linux-next.orig/mm/page_alloc.c     2011-04-28 21:16:16.000000000 +0800
    > +++ linux-next/mm/page_alloc.c  2011-04-28 21:16:18.000000000 +0800
    > @@ -1888,9 +1888,8 @@ __alloc_pages_direct_reclaim(gfp_t gfp_m
    >        nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone,
    >        int migratetype, unsigned long *did_some_progress)
    >  {
    > -       struct page *page = NULL;
    > +       struct page *page;
    >        struct reclaim_state reclaim_state;
    > -       bool drained = false;
    >
    >        cond_resched();
    >
    > @@ -1912,22 +1911,12 @@ __alloc_pages_direct_reclaim(gfp_t gfp_m
    >        if (unlikely(!(*did_some_progress)))
    >                return NULL;
    >
    > -retry:
    > +       alloc_flags |= ALLOC_HARDER;
    > +
    >        page = get_page_from_freelist(gfp_mask, nodemask, order,
    >                                        zonelist, high_zoneidx,
    >                                        alloc_flags, preferred_zone,
    >                                        migratetype);
    > -
    > -       /*
    > -        * If an allocation failed after direct reclaim, it could be because
    > -        * pages are pinned on the per-cpu lists. Drain them and try again
    > -        */
    > -       if (!page && !drained) {
    > -               drain_all_pages();
    > -               drained = true;
    > -               goto retry;
    > -       }
    > -
    >        return page;
    >  }
    >
    >



    --
    Regards
    dave
    \
     
     \ /
      Last update: 2011-05-04 03:59    [W:0.056 / U:92.580 seconds]
    ©2003-2016 Jasper Spaans. hosted at Digital OceanAdvertise on this site