On 12/18/18 2:51 PM, Mel Gorman wrote: > On Tue, Dec 18, 2018 at 01:36:42PM +0100, Vlastimil Babka wrote: >> 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". >> > > That gap is somewhat static though so I'm a bit more wary of it. However, Well, watermark boost is dynamic, but based on allocations stealing from other migratetypes, not reflecting compaction chances of success. > the check in __isolate_free_page looks too agressive. We isolate in > units of COMPACT_CLUSTER_MAX yet the watermark check there is based on > the allocation request. That means for THP that we check if 512 pages > can be allocated when only somewhere between 1 and 32 is needed for that > compaction cycle to complete. Adjusting that might be more appropriate? AFAIU the code in __isolate_free_page() reflects that if there's less than 512 free pages gap, we might form a high-order page for THP but won't be able to allocate it afterwards due to watermark.
