[infinispan-issues] [JBoss JIRA] Created: (ISPN-350) Improve evenness of distribution

[Manik Surtani (JIRA)]

Improve evenness of distribution
--------------------------------

                 Key: ISPN-350
                 URL: https://jira.jboss.org/jira/browse/ISPN-350
             Project: Infinispan
          Issue Type: Feature Request
          Components: Distributed Cache
    Affects Versions: 4.0.0.CR4
            Reporter: Manik Surtani
            Assignee: Manik Surtani
             Fix For: 4.1.0.BETA1, 4.1.0.Final


On this 8-node run, compare the stats measured on each node when using Infinispan:

2010-02-12 07:03:59,123 [main] INFO  [org.cachebench.stages.WebSessionBenchmarkStage] On slave 4 we had 9,739 requests per second
2010-02-12 07:03:59,123 [main] INFO  [org.cachebench.stages.WebSessionBenchmarkStage] On slave 7 we had 3,634 requests per second
2010-02-12 07:03:59,123 [main] INFO  [org.cachebench.stages.WebSessionBenchmarkStage] On slave 1 we had 2,966 requests per second
2010-02-12 07:03:59,123 [main] INFO  [org.cachebench.stages.WebSessionBenchmarkStage] On slave 3 we had 2,935 requests per second
2010-02-12 07:03:59,124 [main] INFO  [org.cachebench.stages.WebSessionBenchmarkStage] On slave 2 we had 2,930 requests per second
2010-02-12 07:03:59,124 [main] INFO  [org.cachebench.stages.WebSessionBenchmarkStage] On slave 5 we had 2,925 requests per second
2010-02-12 07:03:59,124 [main] INFO  [org.cachebench.stages.WebSessionBenchmarkStage] On slave 6 we had 2,922 requests per second
2010-02-12 07:03:59,124 [main] INFO  [org.cachebench.stages.WebSessionBenchmarkStage] On slave 0 we had 2,910 requests per second

against <a popular proprietary product>:

2010-02-12 06:43:44,747 [main] INFO  [org.cachebench.stages.WebSessionBenchmarkStage] On slave 1 we had 12,188 requests per second
2010-02-12 06:43:44,748 [main] INFO  [org.cachebench.stages.WebSessionBenchmarkStage] On slave 7 we had 12,001 requests per second
2010-02-12 06:43:44,748 [main] INFO  [org.cachebench.stages.WebSessionBenchmarkStage] On slave 0 we had 11,978 requests per second
2010-02-12 06:43:44,748 [main] INFO  [org.cachebench.stages.WebSessionBenchmarkStage] On slave 5 we had 11,967 requests per second
2010-02-12 06:43:44,748 [main] INFO  [org.cachebench.stages.WebSessionBenchmarkStage] On slave 3 we had 11,994 requests per second
2010-02-12 06:43:44,749 [main] INFO  [org.cachebench.stages.WebSessionBenchmarkStage] On slave 6 we had 11,961 requests per second
2010-02-12 06:43:44,749 [main] INFO  [org.cachebench.stages.WebSessionBenchmarkStage] On slave 4 we had 11,983 requests per second
2010-02-12 06:43:44,749 [main] INFO  [org.cachebench.stages.WebSessionBenchmarkStage] On slave 2 we had 11,814 requests per second

This is due to the fact that keys are inequally distributed and some nodes have to deal with greater load than others.

We need to revisit the bit spreaders we use as well as the way we place nodes on a hash wheel and try and improve on this.  Virtual nodes may help as well, although that adds significant complexity to the distribution implementation.

-- 
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators: https://jira.jboss.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira