[jira] [Commented] (FLINK-2715) Benchmark Triangle Count methods

["Greg Hogan (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/FLINK-2715?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15254051#comment-15254051 ] 

Greg Hogan commented on FLINK-2715:
-----------------------------------

I happened to include some triangle benchmarks in the pull request for FLINK-3768. Adding a flag for {{GSATriangleCount}} was simple. Testing on an AWS ec2.8xlarge (36 vcores, 60 GiB), what took {{TriangleListing}} 5s and {{TriangleEnumerator}} 7s crashed Flink at 6m3s with {{GSATriangleCount}}. Orkut and Friendster are 100x and 2000x larger than RMat s16e16 so running these benchmarks with GSA or SG does not look feasible with the current algorithms.

I do agree that for overlapping algorithms in Gelly it would be nice to present users a performance comparison. Thoughts, [~vkalavri]?

> Benchmark Triangle Count methods
> --------------------------------
>
>                 Key: FLINK-2715
>                 URL: https://issues.apache.org/jira/browse/FLINK-2715
>             Project: Flink
>          Issue Type: Task
>          Components: Gelly
>    Affects Versions: 0.10.0
>            Reporter: Andra Lungu
>            Priority: Minor
>              Labels: starter
>
> Once FLINK-2714 is addressed, it would be nice to have a set of benchmarks that test the efficiency of the DataSet, GSA and vertex-centric versions. 
> This means running the three examples on a cluster environment using various graph DataSets. For instance, SNAP's Orkut and Friendster networks
> (https://snap.stanford.edu/data/).    
> The results produced by the experiments should then be reported in the Gelly docs.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)