[
https://issues.apache.org/jira/browse/STORM-1731?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15258267#comment-15258267
]
ASF GitHub Bot commented on STORM-1731:
---------------------------------------
Github user HeartSaVioR commented on the pull request:
https://github.com/apache/storm/pull/1362#issuecomment-214781229
I've done performance comparison test with ThroughputVsLatency with one
machine.
* disruptor batching size is set to default (100 / timeout 1ms)
throughput | version | min | max | mean | stddev | 99% | 99.9%
---- | ---- | ---- | ---- | ---- | ---- | ---- | ----
10000 | before patch | 9,437,184 | 23,609,343 | 16,875,865.63 |
1,342,051.21 | 19,611,647 | 20,987,903
10000 | STORM-1731 | 9,273,344 | 23,609,343 | 16,639,346.99 | 1,401,096.42
| 19,333,119 | 20,660,223
22000 | before patch | 9,134,080 | 157,155,327 | 18,968,494.02 |
13,125,204.89 | 95,551,487 | 27,729,663
22000 | STORM-1731 | 8,871,936 | 27,869,183 | 16,865,395.47 | 1,306,652.91
| 19,660,799 | 21,233,663
At throughput 22,000, latency increases steeply from version 'before
patch', but latency increases relatively really small from version 'STORM-1731'.
I will go on tests with throughput 25000, 30000 and post the results, too.
> Avoid looking up debug / backpressure enable flags within critical path
> -----------------------------------------------------------------------
>
> Key: STORM-1731
> URL: https://issues.apache.org/jira/browse/STORM-1731
> Project: Apache Storm
> Issue Type: Bug
> Components: storm-core
> Affects Versions: 1.0.0
> Reporter: Jungtaek Lim
> Assignee: Jungtaek Lim
> Priority: Critical
>
> While profiling the result of STORM-1729, I also found that there're many
> places in critical path which look up the value of flags which are not
> updated dynamically.
> ("get from map" is on top 5 from each spout / bolt thread.)
> This should be fixed.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
