From: Mel Gorman <mgorman@suse.de> To: Stable <stable@vger.kernel.org> Cc: Linux-MM <linux-mm@kvack.org>, LKML <linux-kernel@vger.kernel.org>, Mel Gorman <mgorman@suse.de> Subject: [PATCH 08/34] vmscan: reduce wind up shrinker->nr when shrinker can't do work Date: Mon, 23 Jul 2012 14:38:21 +0100 [thread overview] Message-ID: <1343050727-3045-9-git-send-email-mgorman@suse.de> (raw) In-Reply-To: <1343050727-3045-1-git-send-email-mgorman@suse.de> From: Dave Chinner <dchinner@redhat.com> commit 3567b59aa80ac4417002bf58e35dce5c777d4164 upstream. Stable note: Not tracked in Bugzilla. This patch reduces excessive reclaim of slab objects reducing the amount of information that has to be brought back in from disk. The third and fourth paragram in the series describes the impact. When a shrinker returns -1 to shrink_slab() to indicate it cannot do any work given the current memory reclaim requirements, it adds the entire total_scan count to shrinker->nr. The idea behind this is that when the shrinker is next called and can do work, it will do the work of the previously aborted shrinker call as well. However, if a filesystem is doing lots of allocation with GFP_NOFS set, then we get many, many more aborts from the shrinkers than we do successful calls. The result is that shrinker->nr winds up to it's maximum permissible value (twice the current cache size) and then when the next shrinker call that can do work is issued, it has enough scan count built up to free the entire cache twice over. This manifests itself in the cache going from full to empty in a matter of seconds, even when only a small part of the cache is needed to be emptied to free sufficient memory. Under metadata intensive workloads on ext4 and XFS, I'm seeing the VFS caches increase memory consumption up to 75% of memory (no page cache pressure) over a period of 30-60s, and then the shrinker empties them down to zero in the space of 2-3s. This cycle repeats over and over again, with the shrinker completely trashing the inode and dentry caches every minute or so the workload continues. This behaviour was made obvious by the shrink_slab tracepoints added earlier in the series, and made worse by the patch that corrected the concurrent accounting of shrinker->nr. To avoid this problem, stop repeated small increments of the total scan value from winding shrinker->nr up to a value that can cause the entire cache to be freed. We still need to allow it to wind up, so use the delta as the "large scan" threshold check - if the delta is more than a quarter of the entire cache size, then it is a large scan and allowed to cause lots of windup because we are clearly needing to free lots of memory. If it isn't a large scan then limit the total scan to half the size of the cache so that windup never increases to consume the whole cache. Reducing the total scan limit further does not allow enough wind-up to maintain the current levels of performance, whilst a higher threshold does not prevent the windup from freeing the entire cache under sustained workloads. Signed-off-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Mel Gorman <mgorman@suse.de> --- mm/vmscan.c | 15 +++++++++++++++ 1 file changed, 15 insertions(+) diff --git a/mm/vmscan.c b/mm/vmscan.c index 31b551e..8ca1cd5 100644 --- a/mm/vmscan.c +++ b/mm/vmscan.c @@ -277,6 +277,21 @@ unsigned long shrink_slab(struct shrink_control *shrink, } /* + * We need to avoid excessive windup on filesystem shrinkers + * due to large numbers of GFP_NOFS allocations causing the + * shrinkers to return -1 all the time. This results in a large + * nr being built up so when a shrink that can do some work + * comes along it empties the entire cache due to nr >>> + * max_pass. This is bad for sustaining a working set in + * memory. + * + * Hence only allow the shrinker to scan the entire cache when + * a large delta change is calculated directly. + */ + if (delta < max_pass / 4) + total_scan = min(total_scan, max_pass / 2); + + /* * Avoid risking looping forever due to too large nr value: * never try to free more than twice the estimate number of * freeable entries. -- 1.7.9.2
WARNING: multiple messages have this Message-ID (diff)
From: Mel Gorman <mgorman@suse.de> To: Stable <stable@vger.kernel.org> Cc: Linux-MM <linux-mm@kvack.org>, LKML <linux-kernel@vger.kernel.org>, Mel Gorman <mgorman@suse.de> Subject: [PATCH 08/34] vmscan: reduce wind up shrinker->nr when shrinker can't do work Date: Mon, 23 Jul 2012 14:38:21 +0100 [thread overview] Message-ID: <1343050727-3045-9-git-send-email-mgorman@suse.de> (raw) In-Reply-To: <1343050727-3045-1-git-send-email-mgorman@suse.de> From: Dave Chinner <dchinner@redhat.com> commit 3567b59aa80ac4417002bf58e35dce5c777d4164 upstream. Stable note: Not tracked in Bugzilla. This patch reduces excessive reclaim of slab objects reducing the amount of information that has to be brought back in from disk. The third and fourth paragram in the series describes the impact. When a shrinker returns -1 to shrink_slab() to indicate it cannot do any work given the current memory reclaim requirements, it adds the entire total_scan count to shrinker->nr. The idea behind this is that when the shrinker is next called and can do work, it will do the work of the previously aborted shrinker call as well. However, if a filesystem is doing lots of allocation with GFP_NOFS set, then we get many, many more aborts from the shrinkers than we do successful calls. The result is that shrinker->nr winds up to it's maximum permissible value (twice the current cache size) and then when the next shrinker call that can do work is issued, it has enough scan count built up to free the entire cache twice over. This manifests itself in the cache going from full to empty in a matter of seconds, even when only a small part of the cache is needed to be emptied to free sufficient memory. Under metadata intensive workloads on ext4 and XFS, I'm seeing the VFS caches increase memory consumption up to 75% of memory (no page cache pressure) over a period of 30-60s, and then the shrinker empties them down to zero in the space of 2-3s. This cycle repeats over and over again, with the shrinker completely trashing the inode and dentry caches every minute or so the workload continues. This behaviour was made obvious by the shrink_slab tracepoints added earlier in the series, and made worse by the patch that corrected the concurrent accounting of shrinker->nr. To avoid this problem, stop repeated small increments of the total scan value from winding shrinker->nr up to a value that can cause the entire cache to be freed. We still need to allow it to wind up, so use the delta as the "large scan" threshold check - if the delta is more than a quarter of the entire cache size, then it is a large scan and allowed to cause lots of windup because we are clearly needing to free lots of memory. If it isn't a large scan then limit the total scan to half the size of the cache so that windup never increases to consume the whole cache. Reducing the total scan limit further does not allow enough wind-up to maintain the current levels of performance, whilst a higher threshold does not prevent the windup from freeing the entire cache under sustained workloads. Signed-off-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Mel Gorman <mgorman@suse.de> --- mm/vmscan.c | 15 +++++++++++++++ 1 file changed, 15 insertions(+) diff --git a/mm/vmscan.c b/mm/vmscan.c index 31b551e..8ca1cd5 100644 --- a/mm/vmscan.c +++ b/mm/vmscan.c @@ -277,6 +277,21 @@ unsigned long shrink_slab(struct shrink_control *shrink, } /* + * We need to avoid excessive windup on filesystem shrinkers + * due to large numbers of GFP_NOFS allocations causing the + * shrinkers to return -1 all the time. This results in a large + * nr being built up so when a shrink that can do some work + * comes along it empties the entire cache due to nr >>> + * max_pass. This is bad for sustaining a working set in + * memory. + * + * Hence only allow the shrinker to scan the entire cache when + * a large delta change is calculated directly. + */ + if (delta < max_pass / 4) + total_scan = min(total_scan, max_pass / 2); + + /* * Avoid risking looping forever due to too large nr value: * never try to free more than twice the estimate number of * freeable entries. -- 1.7.9.2 -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@kvack.org. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
next prev parent reply other threads:[~2012-07-23 13:39 UTC|newest] Thread overview: 119+ messages / expand[flat|nested] mbox.gz Atom feed top 2012-07-23 13:38 [PATCH 00/34] Memory management performance backports for -stable V2 Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-23 13:38 ` [PATCH 01/34] mm: vmstat: cache align vm_stat Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-23 13:38 ` [PATCH 02/34] mm: memory hotplug: Check if pages are correctly reserved on a per-section basis Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-23 13:38 ` [PATCH 03/34] mm: Reduce the amount of work done when updating min_free_kbytes Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-24 22:47 ` Greg KH 2012-07-24 22:47 ` Greg KH 2012-07-25 7:57 ` Mel Gorman 2012-07-25 7:57 ` Mel Gorman 2012-07-23 13:38 ` [PATCH 04/34] mm: vmscan: fix force-scanning small targets without swap Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-23 13:38 ` [PATCH 05/34] vmscan: clear ZONE_CONGESTED for zone with good watermark Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-23 13:38 ` [PATCH 06/34] vmscan: add shrink_slab tracepoints Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-23 13:38 ` [PATCH 07/34] vmscan: shrinker->nr updates race and go wrong Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-23 13:38 ` Mel Gorman [this message] 2012-07-23 13:38 ` [PATCH 08/34] vmscan: reduce wind up shrinker->nr when shrinker can't do work Mel Gorman 2012-07-23 13:38 ` [PATCH 09/34] mm: limit direct reclaim for higher order allocations Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-23 13:38 ` [PATCH 10/34] mm: Abort reclaim/compaction if compaction can proceed Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-23 13:38 ` [PATCH 11/34] mm: compaction: trivial clean up in acct_isolated() Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-23 13:38 ` [PATCH 12/34] mm: change isolate mode from #define to bitwise type Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-23 13:38 ` [PATCH 13/34] mm: compaction: make isolate_lru_page() filter-aware Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-23 13:38 ` [PATCH 14/34] mm: zone_reclaim: " Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-23 13:38 ` [PATCH 15/34] mm: migration: clean up unmap_and_move() Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-25 15:45 ` Greg KH 2012-07-25 15:45 ` Greg KH 2012-07-25 16:04 ` Mel Gorman 2012-07-25 16:04 ` Mel Gorman 2012-07-25 18:03 ` Greg KH 2012-07-25 18:03 ` Greg KH 2012-07-23 13:38 ` [PATCH 16/34] mm: compaction: Allow compaction to isolate dirty pages Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-25 15:47 ` Greg KH 2012-07-25 15:47 ` Greg KH 2012-07-25 16:07 ` Mel Gorman 2012-07-25 16:07 ` Mel Gorman 2012-07-23 13:38 ` [PATCH 17/34] mm: compaction: Determine if dirty pages can be migrated without blocking within ->migratepage Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-23 13:38 ` [PATCH 18/34] mm: page allocator: Do not call direct reclaim for THP allocations while compaction is deferred Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-23 13:38 ` [PATCH 19/34] mm: compaction: make isolate_lru_page() filter-aware again Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-23 13:38 ` [PATCH 20/34] kswapd: avoid unnecessary rebalance after an unsuccessful balancing Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-23 13:38 ` [PATCH 21/34] kswapd: assign new_order and new_classzone_idx after wakeup in sleeping Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-23 13:38 ` [PATCH 22/34] mm: compaction: Introduce sync-light migration for use by compaction Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-23 13:38 ` [PATCH 23/34] mm: vmscan: When reclaiming for compaction, ensure there are sufficient free pages available Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-23 13:38 ` [PATCH 24/34] mm: vmscan: Do not OOM if aborting reclaim to start compaction Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-23 13:38 ` [PATCH 25/34] mm: vmscan: Check if reclaim should really abort even if compaction_ready() is true for one zone Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-25 19:51 ` Greg KH 2012-07-25 19:51 ` Greg KH 2012-07-23 13:38 ` [PATCH 26/34] vmscan: promote shared file mapped pages Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-23 13:38 ` [PATCH 27/34] vmscan: activate executable pages after first usage Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-23 13:38 ` [PATCH 28/34] mm/vmscan.c: consider swap space when deciding whether to continue reclaim Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-23 13:38 ` [PATCH 29/34] mm: test PageSwapBacked in lumpy reclaim Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-23 13:38 ` [PATCH 30/34] mm: vmscan: Do not force kswapd to scan small targets Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-25 19:59 ` Greg KH 2012-07-25 19:59 ` Greg KH 2012-07-25 21:35 ` Mel Gorman 2012-07-25 21:35 ` Mel Gorman 2012-07-25 21:44 ` Greg KH 2012-07-25 21:44 ` Greg KH 2012-07-23 13:38 ` [PATCH 31/34] cpusets: avoid looping when storing to mems_allowed if one node remains set Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-23 13:38 ` [PATCH 32/34] cpusets: stall when updating mems_allowed for mempolicy or disjoint nodemask Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-23 13:38 ` [PATCH 33/34] cpuset: mm: Reduce large amounts of memory barrier related damage v3 Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-23 13:38 ` [PATCH 34/34] mm/hugetlb: fix warning in alloc_huge_page/dequeue_huge_page_vma Mel Gorman 2012-07-23 13:38 ` Mel Gorman 2012-07-24 5:58 ` [PATCH 00/34] Memory management performance backports for -stable V2 Mike Galbraith 2012-07-24 5:58 ` Mike Galbraith 2012-07-24 8:10 ` Mel Gorman 2012-07-24 8:10 ` Mel Gorman 2012-07-24 13:18 ` Hillf Danton 2012-07-24 13:18 ` Hillf Danton 2012-07-24 13:27 ` Mel Gorman 2012-07-24 13:27 ` Mel Gorman 2012-07-24 13:34 ` Hillf Danton 2012-07-24 13:34 ` Hillf Danton 2012-07-24 13:53 ` Mel Gorman 2012-07-24 13:53 ` Mel Gorman 2012-07-24 14:11 ` Hillf Danton 2012-07-24 14:11 ` Hillf Danton 2012-07-24 13:52 ` Mike Galbraith 2012-07-24 13:52 ` Mike Galbraith 2012-07-24 14:18 ` Hillf Danton 2012-07-24 14:18 ` Hillf Danton 2012-07-24 14:41 ` Mike Galbraith 2012-07-24 14:41 ` Mike Galbraith 2012-07-25 22:30 ` Greg KH 2012-07-25 22:30 ` Greg KH 2012-07-25 22:48 ` Mel Gorman 2012-07-25 22:48 ` Mel Gorman 2012-07-30 1:13 ` Ben Hutchings -- strict thread matches above, loose matches on Subject: below -- 2012-07-19 14:36 [PATCH 00/34] Memory management performance backports for -stable Mel Gorman 2012-07-19 14:36 ` [PATCH 08/34] vmscan: reduce wind up shrinker->nr when shrinker can't do work Mel Gorman 2012-07-19 14:36 ` Mel Gorman
Reply instructions: You may reply publicly to this message via plain-text email using any one of the following methods: * Save the following mbox file, import it into your mail client, and reply-to-all from there: mbox Avoid top-posting and favor interleaved quoting: https://en.wikipedia.org/wiki/Posting_style#Interleaved_style * Reply using the --to, --cc, and --in-reply-to switches of git-send-email(1): git send-email \ --in-reply-to=1343050727-3045-9-git-send-email-mgorman@suse.de \ --to=mgorman@suse.de \ --cc=linux-kernel@vger.kernel.org \ --cc=linux-mm@kvack.org \ --cc=stable@vger.kernel.org \ /path/to/YOUR_REPLY https://kernel.org/pub/software/scm/git/docs/git-send-email.html * If your mail client supports setting the In-Reply-To header via mailto: links, try the mailto: linkBe sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes, see mirroring instructions on how to clone and mirror all data and code used by this external index.