Hi, this is based on top of [1]. The first patch continues in the direction of moving some decisions to zones rather than nodes. In this case it is the NR_ISOLATED* counters which I believe need to be zone aware as well. See patch 1 for more information why. The second path builds on top of that and tries to address the problem which has been reported by Tetsuo several times already. In the current implementation we can loop deep in the reclaim path without any effective way out to re-evaluate our decisions about the reclaim retries. Patch 2 says more about that but in principle we should locate retry logic as high in the allocator chain as possible and so we should get rid of any unbound retry loops inside the reclaim. This is what the patch does. I am sending this as an RFC because I am not yet sure this is the best forward. My testing shows that the system behaves sanely. Thoughts, comments? [1] http://lkml.kernel.org/r/20170117103702.28542-1-mhocko@xxxxxxxxxx Michal Hocko (2): mm, vmscan: account the number of isolated pages per zone mm, vmscan: do not loop on too_many_isolated for ever include/linux/mmzone.h | 4 +-- mm/compaction.c | 16 ++++----- mm/khugepaged.c | 4 +-- mm/memory_hotplug.c | 2 +- mm/migrate.c | 4 +-- mm/page_alloc.c | 14 ++++---- mm/vmscan.c | 93 ++++++++++++++++++++++++++++++++------------------ mm/vmstat.c | 4 +-- 8 files changed, 82 insertions(+), 59 deletions(-) -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>