[jira] [Commented] (GEODE-6190) Investigate reproducability of ReplicatedPutBenchmark

["Dan Smith (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/GEODE-6190?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16718291#comment-16718291 ] 

Dan Smith commented on GEODE-6190:
----------------------------------

Here is a chart of the throughput of two back to back runs of ReplicatedPutBenchmark in AWS with this setup. As you can see, the runs have a different throughput despite being on the same hardware and same revisions. We see similar results running on dedicated physical machines, so this is likely not due to AWS.

Benchmark SHA: e78aa2858e - feature/fixed_48_threads (0ee9ec6e0c57 + a commit to set the # of threads to 48)
Geode SHA: rel/v1.7.0
Infrastructure: 4 c5d.9xlarge AWS instances, in their own placement group, and marked as dedicated instances. In the default VPC for the us-west region. Ubuntu 16.04


 !two-runs-of-ReplicatedPutBenchmark.png! 

> Investigate reproducability of ReplicatedPutBenchmark
> -----------------------------------------------------
>
>                 Key: GEODE-6190
>                 URL: https://issues.apache.org/jira/browse/GEODE-6190
>             Project: Geode
>          Issue Type: Task
>          Components: benchmarks
>            Reporter: Dan Smith
>            Priority: Major
>         Attachments: two-runs-of-ReplicatedPutBenchmark.png
>
>
> When running the replicated put benchmark on dedicated hardware or in AWS with cluster placement groups, we can observe a clear difference in the throughput of the ReplicatedPutBenchmark between runs. See the attached charts.
> Investigate if this variance still reproduces after GEODE-6172 and GEODE-6148 and implemented. If so, investigate why we are seeing varying throughput from this test, and eliminate the source of variance.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)