Hi Alex,
On 8/20/19 5:48 AM, Alex Shi wrote:
In some data center, containers are used widely to deploy different kind
of services, then multiple memcgs share per node pgdat->lru_lock which
cause heavy lock contentions when doing lru operation.
On my 2 socket * 6 cores E5-2630 platform, 24 containers run aim9
simultaneously with mmtests' config:
# AIM9
export AIM9_TESTTIME=180
export AIM9_TESTLIST=page_test,brk_test
perf lock report show much contentions on lru_lock in 20 second snapshot:
Name acquired contended avg wait (ns) total wait (ns) max wait (ns) min wait (ns)
&(ptlock_ptr(pag... 22 0 0 0 0 0
...
&(&pgdat->lru_lo... 9 7 12728 89096 26656 1597
This is system-wide right, not per container? Even per container, 89 usec isn't much contention over 20 seconds. You may want to give this a try:
https://git.kernel.org/pub/scm/linux/kernel/git/wfg/vm-scalability.git/tree/case-lru-file-readtwice
It's also synthetic but it stresses lru_lock more than just anon alloc/free. It hits the page activate path, which is where we see this lock in our database, and if enough memory is configured lru_lock also gets stressed during reclaim, similar to [1].
It'd be better though, as Michal suggests, to use the real workload that's causing problems. Where are you seeing contention?
With this patch series, lruvec->lru_lock show no contentions
&(&lruvec->lru_l... 8 0 0 0 0 0
and aim9 page_test/brk_test performance increased 5%~50%.
Where does the 50% number come in? The numbers below seem to only show ~4% boost.
BTW, Detailed results in aim9-pft.compare.log if needed,
All containers data are increased and pretty steady.
$for i in Max Min Hmean Stddev CoeffVar BHmean-50 BHmean-95 BHmean-99; do echo "========= $i page_test ============"; cat aim9-pft.compare.log | grep "^$i.*page_test" | awk 'BEGIN {a=b=0;} { a+=$3; b+=$6 } END { print "5.3-rc4 " a/24; print "5.3-rc4+lru_lock " b/24}' ; done
========= Max page_test ============
5.3-rc4 34729.6
5.3-rc4+lru_lock 36128.3
========= Min page_test ============
5.3-rc4 33644.2
5.3-rc4+lru_lock 35349.7
========= Hmean page_test ============
5.3-rc4 34355.4
5.3-rc4+lru_lock 35810.9
========= Stddev page_test ============
5.3-rc4 319.757
5.3-rc4+lru_lock 223.324
========= CoeffVar page_test ============
5.3-rc4 0.93125
5.3-rc4+lru_lock 0.623333
========= BHmean-50 page_test ============
5.3-rc4 34579.2
5.3-rc4+lru_lock 35977.1
========= BHmean-95 page_test ============
5.3-rc4 34421.7
5.3-rc4+lru_lock 35853.6
========= BHmean-99 page_test ============
5.3-rc4 34421.7
5.3-rc4+lru_lock 35853.6
$for i in Max Min Hmean Stddev CoeffVar BHmean-50 BHmean-95 BHmean-99; do echo "========= $i brk_test ============"; cat aim9-pft.compare.log | grep "^$i.*brk_test" | awk 'BEGIN {a=b=0;} { a+=$3; b+=$6 } END { print "5.3-rc4 " a/24; print "5.3-rc4+lru_lock " b/24}' ; done
========= Max brk_test ============
5.3-rc4 96647.7
5.3-rc4+lru_lock 98960.3
========= Min brk_test ============
5.3-rc4 91800.8
5.3-rc4+lru_lock 96817.6
========= Hmean brk_test ============
5.3-rc4 95470
5.3-rc4+lru_lock 97769.6
========= Stddev brk_test ============
5.3-rc4 1253.52
5.3-rc4+lru_lock 596.593
========= CoeffVar brk_test ============
5.3-rc4 1.31375
5.3-rc4+lru_lock 0.609583
========= BHmean-50 brk_test ============
5.3-rc4 96141.4
5.3-rc4+lru_lock 98194
========= BHmean-95 brk_test ============
5.3-rc4 95818.5
5.3-rc4+lru_lock 97857.2
========= BHmean-99 brk_test ============
5.3-rc4 95818.5
5.3-rc4+lru_lock 97857.2
[1] https://lore.kernel.org/linux-mm/CABdVr8R2y9B+2zzSAT_Ve=BQCa+F+E9_kVH+C28DGpkeQitiog@xxxxxxxxxxxxxx/
