Hi,
On 02/06/2017 11:17 AM, Mel Gorman wrote:
On Sun, Feb 05, 2017 at 12:46:20PM +0000, Shantanu Goel wrote:
On 4.9.7 kswapd is failing to wake up kcompactd due to a mismatch in the zone balance check between balance_pgdat() and prepare_kswapd_sleep(). balance_pgdat() returns as soon as a single zone satisfies the allocation but prepare_kswapd_sleep() requires all zones to do the same. This causes prepare_kswapd_sleep() to never succeed except in the order == 0 case and consequently, wakeup_kcompactd() is never called. On my machine prior to apply this patch, the state of compaction from /proc/vmstat looked this way after a day and a half of uptime:
compact_migrate_scanned 240496
compact_free_scanned 76238632
compact_isolated 123472
compact_stall 1791
compact_fail 29
compact_success 1762
compact_daemon_wake 0
After applying the patch and about 10 hours of uptime the state looks like this:
compact_migrate_scanned 59927299
compact_free_scanned 2021075136
compact_isolated 640926
compact_stall 4
compact_fail 2
compact_success 2
compact_daemon_wake 5160
This should be in the changelog of the patch itself and the patch should
be inline instead of being an attachment.
Will do and resubmit in a separate email.
Thanks,
Shantanu
From 46f2e4b02ac263bf50d69cdab3bcbd7bcdea7415 Mon Sep 17 00:00:00 2001
From: Shantanu Goel <sgoel01@xxxxxxxxx>
Date: Sat, 4 Feb 2017 19:07:53 -0500
Subject: [PATCH] vmscan: fix zone balance check in prepare_kswapd_sleep
The check in prepare_kswapd_sleep needs to match the one in balance_pgdat
since the latter will return as soon as any one of the zones in the
classzone is above the watermark. This is specially important for
higher order allocations since balance_pgdat will typically reset
the order to zero relying on compaction to create the higher order
pages. Without this patch, prepare_kswapd_sleep fails to wake up
kcompactd since the zone balance check fails.
Signed-off-by: Shantanu Goel <sgoel01@xxxxxxxxx>
I don't recall specifically why I made that change but I've no objections
to the patch so;
Acked-by: Mel Gorman <mgorman@xxxxxxxxxxxxxxxxxxx>
However, note that there is a slight risk that kswapd will sleep for a
short interval early due to a very small zone such as ZONE_DMA. If this
is a general problem then it'll manifest as less kswapd reclaim and more
direct reclaim. If it turns out this is an issue then a revert will not
be the right fix. Instead, all the checks for zone_balance will need to
account for the only balanced zone being a tiny percentage of memory in
the node.
I see your point. Perhaps we can introduce a constraint that
ensures the balanced zones constitute say 1/4 or 1/2 of
memory in the classzone? I believe there used to be such
a constraint at one time for higher order allocations.
Thanks,
Shantanu
--
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>
