Skip to content

RFC vmscan: AIM7 scalability improvement

KOSAKI Motohiro linux-mm lkml 2 versions – In review

Get series with b4
b4 am [email protected]

Versions newest first · linked by author and title

  1. v2 RFC Latest8 patches 4 repliesIn review
  2. v1 RFC7 patches 11 repliesSuperseded

Patches in v2 No applies recorded yet

#Subject and filesSizeRepliesApplied
1/8Replace page_mapping_inuse() with page_mapped()mm/vmscan.c+2 -230 replies
2/8Introduce __page_check_addressmm/rmap.c+49 -130 replies
3/8VM_LOCKED check don't need pte lockmm/rmap.c+6 -70 replies
4/8Replace page_referenced() with wipe_page_reference()include/linux/ksm.h include/linux/rmap.h mm/ksm.c +2 more+112 -902 replies
5/8Don't deactivate the page if trylock_page() is failed.mm/rmap.c mm/vmscan.c+17 -30 replies
6/8wipe_page_reference return SWAP_AGAIN if VM pressulre is low and lock contention is detected.include/linux/rmap.h mm/rmap.c mm/vmscan.c+17 -40 replies
7/8Try to mark PG_mlocked if wipe_page_reference find VM_LOCKED vmamm/rmap.c mm/vmscan.c+16 -00 replies
8/8Don't deactivate many touched pageinclude/linux/rmap.h mm/ksm.c mm/rmap.c+41 -01 reply

Reviews trailers per patch

All 8 patches have a Reviewed-by or Acked-by.

PatchRik van RielJohannes Weiner
1/8R-b R-b
2/8R-b
3/8R-b
4/8R-b
5/8R-b
6/8A-b
7/8R-b
8/8R-b

Trailers given in replies are filled; those carried in the patches from earlier versions are outlined with dashes.

Read the v2 thread

Cover letter [RFC][PATCH v2 0/8] vmscan: AIM7 scalability improvement

Larry Woodman reported current VM has big performance regression when AIM7 benchmark. It use very much processes and fork/exit/page-fault frequently. (i.e. it makes serious lock contention of ptelock and anon_vma-lock.)

At 2.6.28, we removed calc_reclaim_mapped() and it made vmscan, then vmscan always call page_referenced() although VM pressure is low. It increased lock contention more, unfortunately.

Larry, can you please try this patch series on your big box?