[
https://issues.apache.org/jira/browse/HBASE-14268?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14730483#comment-14730483
]
Hiroshi Ikeda commented on HBASE-14268:
---------------------------------------
For performance tests in much GC, I multiplied LOOP_COUNT by 10 or 100 and run
the test application with the VM option -verbose:gc. The new KeyLocker works
well. The old KeyLocker immediately drops released locks and GC happens
frequently.
new KeyLocker, LOOP_COUNT=1,000,000
{code}
[GC 33280K->789K(124928K), 0.0018632 secs]
[GC 34069K->733K(124928K), 0.0015329 secs]
[GC 34013K->733K(124928K), 0.0012143 secs]
[GC 34013K->733K(158208K), 0.0012232 secs]
2113838581
{code}
new KeyLocker, LOOP_COUNT=10,000,000
{code}
[GC 33280K->757K(124928K), 0.0021142 secs]
[GC 34037K->693K(124928K), 0.0046123 secs]
[GC 33973K->725K(124928K), 0.0128241 secs]
[GC 34005K->725K(158208K), 0.0012777 secs]
[GC 67285K->693K(158208K), 0.0013967 secs]
[GC 67253K->725K(220672K), 0.0012992 secs]
[GC 133845K->838K(220672K), 0.0014905 secs]
[GC 133958K->870K(353792K), 0.0004786 secs]
19860706850
{code}
old KeyLocker, LOOP_COUNT=1,000,000
{code}
[GC 33280K->1032K(124928K), 0.0018203 secs]
[GC 34312K->789K(124928K), 0.0014084 secs]
[GC 34069K->853K(124928K), 0.0012749 secs]
[GC 34133K->853K(158208K), 0.0012367 secs]
[GC 67413K->853K(158208K), 0.0013319 secs]
[GC 67413K->821K(220672K), 0.0012880 secs]
[GC 133941K->882K(220672K), 0.0015950 secs]
[GC 134002K->882K(354304K), 0.0005267 secs]
[GC 267122K->882K(354304K), 0.0007189 secs]
[GC 267122K->818K(514048K), 0.0087180 secs]
[GC 426802K->882K(514048K), 0.0008262 secs]
[GC 426866K->914K(770048K), 0.0005164 secs]
[GC 682898K->914K(770048K), 0.0005015 secs]
[GC 682898K->914K(780288K), 0.0004968 secs]
[GC 693138K->850K(780288K), 0.0005118 secs]
[GC 693074K->882K(746496K), 0.0005215 secs]
[GC 660338K->914K(716288K), 0.0005057 secs]
[GC 629138K->914K(685568K), 0.0005206 secs]
[GC 599442K->882K(658432K), 0.0061237 secs]
[GC 571250K->882K(630784K), 0.0005043 secs]
[GC 544626K->850K(605696K), 0.0004590 secs]
40161876412
{code}
On the other hand, I found that, in the case of locking and releasing new keys
at all times, the new KeyLocker has less performance. That is because weak
references may require a several GC, and I think that happens especially when
there are too many weak references. I'll append later such a test application
and its result.
> Improve KeyLocker
> -----------------
>
> Key: HBASE-14268
> URL: https://issues.apache.org/jira/browse/HBASE-14268
> Project: HBase
> Issue Type: Improvement
> Components: util
> Reporter: Hiroshi Ikeda
> Assignee: Hiroshi Ikeda
> Priority: Minor
> Fix For: 2.0.0, 1.3.0
>
> Attachments: 14268-V5.patch, HBASE-14268-V2.patch,
> HBASE-14268-V3.patch, HBASE-14268-V4.patch, HBASE-14268-V5.patch,
> HBASE-14268-V5.patch, HBASE-14268-V6.patch, HBASE-14268-V7.patch,
> HBASE-14268-V7.patch, HBASE-14268.patch, KeyLockerPerformance.java
>
>
> 1. In the implementation of {{KeyLocker}} it uses atomic variables inside a
> synchronized block, which doesn't make sense. Moreover, logic inside the
> synchronized block is not trivial so that it makes less performance in heavy
> multi-threaded environment.
> 2. {{KeyLocker}} gives an instance of {{RentrantLock}} which is already
> locked, but it doesn't follow the contract of {{ReentrantLock}} because you
> are not allowed to freely invoke lock/unlock methods under that contract.
> That introduces a potential risk; Whenever you see a variable of the type
> {{RentrantLock}}, you should pay attention to what the included instance is
> coming from.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
