[
https://issues.apache.org/jira/browse/HAMA-761?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13678767#comment-13678767
]
MaoYuan Xian commented on HAMA-761:
-----------------------------------
Since GraphJobMessage only needs VertexId when does the comparation
(VertexValue is useless), I define a special comparator for GraphJobMessage,
which skips the deserialize operation of VertexValue when does the comparatioin.
In real project, size of VertexId:VertextValue is about 1:3, after this
modification, I can gain about 30% performance enhancement.
I push the modifications to
https://github.com/kennethxian/hama/commit/045408e316e9757364fae49fdeafb544b4b2d829
Is is useful, please give some comments.
> Improvement for GraphJobMessage comparation operation
> -----------------------------------------------------
>
> Key: HAMA-761
> URL: https://issues.apache.org/jira/browse/HAMA-761
> Project: Hama
> Issue Type: Improvement
> Components: graph
> Affects Versions: 0.6.1
> Reporter: MaoYuan Xian
> Assignee: MaoYuan Xian
>
> Currently, GraphJobMessage comparation opertion depends on the default
> comparator (org.apache.hadoop.io.WritableComparator used), which has to read
> out all GraphJobMessage message (including vertexId and vertexValue) when
> does the comparation.
> Since lots of comparation operations happen in graph computing, this
> limitation makes a performance downgrade.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
