You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@giraph.apache.org by "Alessandro Presta (JIRA)" <ji...@apache.org> on 2013/02/22 19:26:13 UTC
[jira] [Commented] (GIRAPH-535) Range-partitioning and edge
locality benchmark
[ https://issues.apache.org/jira/browse/GIRAPH-535?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13584509#comment-13584509 ]
Alessandro Presta commented on GIRAPH-535:
------------------------------------------
https://reviews.apache.org/r/9562/
> Range-partitioning and edge locality benchmark
> ----------------------------------------------
>
> Key: GIRAPH-535
> URL: https://issues.apache.org/jira/browse/GIRAPH-535
> Project: Giraph
> Issue Type: New Feature
> Reporter: Alessandro Presta
> Assignee: Alessandro Presta
> Attachments: GIRAPH-535.patch
>
>
> Range-based partitioning can drastically reduce network communication when using a vertex id space where close ids correspond to vertices likely to be connected.
> We currently have an incomplete implementation of range-based partitioning that tries to be very generic (allowing arbitrarily different partition sizes).
> Talking about this with Avery, we thought that for now it's better to add a simpler version (which tries to split as evenly as possible) and leave the current classes there in case someone wants to implement a more complex logic.
> A nice-to-have is also extending the pseudo-random formats to generate a required ratio of partition-local edges, in order to estimate the impact of locality with benchmarks.
--
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