On 12/15/18 12:03 AM, Mel Gorman wrote: > When pageblocks get fragmented, watermarks are artifically boosted to pages > are reclaimed to avoid further fragmentation events. However, compaction > is often either fragmentation-neutral or moving movable pages away from > unmovable/reclaimable pages. As the actual watermarks are preserved, > allow compaction to ignore the boost factor. Right, I should have realized that when reviewing the boost patch. I think it would be useful to do the same change in __compaction_suitable() as well. Compaction has its own "gap". > 1-socket thpscale > 4.20.0-rc6 4.20.0-rc6 > finishscan-v1r4 noboost-v1r4 > Amean fault-both-1 0.00 ( 0.00%) 0.00 * 0.00%* > Amean fault-both-3 3849.90 ( 0.00%) 3753.53 ( 2.50%) > Amean fault-both-5 5054.13 ( 0.00%) 5396.32 ( -6.77%) > Amean fault-both-7 7061.77 ( 0.00%) 7393.46 ( -4.70%) > Amean fault-both-12 11560.59 ( 0.00%) 12155.50 ( -5.15%) > Amean fault-both-18 16120.15 ( 0.00%) 16445.96 ( -2.02%) > Amean fault-both-24 19804.31 ( 0.00%) 20465.03 ( -3.34%) > Amean fault-both-30 25018.73 ( 0.00%) 20813.54 * 16.81%* > Amean fault-both-32 24380.19 ( 0.00%) 22384.02 ( 8.19%) > > The impact on the scan rates is a mixed bag because this patch is very > sensitive to timing and whether the boost was active or not. However, > detailed tracing indicated that failure of migration due to a premature > ENOMEM triggered by watermark checks were eliminated. > > Signed-off-by: Mel Gorman <mgorman@xxxxxxxxxxxxxxxxxxx> > --- > mm/page_alloc.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/mm/page_alloc.c b/mm/page_alloc.c > index 80535cd55a92..c7b80e62bfd9 100644 > --- a/mm/page_alloc.c > +++ b/mm/page_alloc.c > @@ -3043,7 +3043,7 @@ int __isolate_free_page(struct page *page, unsigned int order) > * watermark, because we already know our high-order page > * exists. > */ > - watermark = min_wmark_pages(zone) + (1UL << order); > + watermark = zone->_watermark[WMARK_MIN] + (1UL << order); > if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA)) > return 0; > >
