[ https://issues.apache.org/jira/browse/FLINK-8227?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16299640#comment-16299640 ]
ASF GitHub Bot commented on FLINK-8227: --------------------------------------- Github user dianfu commented on the issue: https://github.com/apache/flink/pull/5142 @dawidwys @StephanEwen Sorry for late response. For question 1 and 2, I have the same thought with @dawidwys and have updated the PR accordingly. For question 3, I think `int` is enough as we currently store `SharedBufferEntry` in a `HashMap` for each `SharedBufferPage`, and the size of `HashMap` is `int`. If we want to support `long`, we should also change `HashMap` to something else. What's your thought? > Optimize the performance of SharedBufferSerializer > -------------------------------------------------- > > Key: FLINK-8227 > URL: https://issues.apache.org/jira/browse/FLINK-8227 > Project: Flink > Issue Type: Bug > Components: CEP > Reporter: Dian Fu > Assignee: Dian Fu > > Currently {{SharedBufferSerializer.serialize()}} will create a HashMap and > put all the {{SharedBufferEntry}} into it. Usually this is not a problem. But > we obverse that in some cases the calculation of hashCode may become the > bottleneck. The performance will decrease as the number of > {{SharedBufferEdge}} increases. For looping pattern {{A*}}, if the number of > {{SharedBufferEntry}} is {{N}}, the number of {{SharedBufferEdge}} is about > {{N * N}}. -- This message was sent by Atlassian JIRA (v6.4.14#64029)