[
https://issues.apache.org/jira/browse/FLINK-1536?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15732898#comment-15732898
]
Ivan Mushketyk edited comment on FLINK-1536 at 12/8/16 6:01 PM:
----------------------------------------------------------------
Do I understand correctly that as the result of this issue we should have an
interface like this:
List<Graph> partition(Graph graph);
namely we will have something that takes an a graph as an input and produces a
list of partitions that we are interested in?
What confuses me is this part of the Gell roadmap
(https://cwiki.apache.org/confluence/display/FLINK/Flink+Gelly):
"Graph Partitioning plays a key role in application parallelization and in
scaling data analysis up. Processes need to evenly be assigned to machines
while maintaining communication costs to a minimum."
Does it mean that partitioning should be a step in other graph processing
algorithms?
>From the same document. What is "hash/random partitioning"?
was (Author: ivan.mushketyk):
Do I understand correctly that as the result of this issue we should have an
interface like this:
List<Graph> partition(Graph graph);
namely we will have something that takes an a graph as an input and produces a
list of partitions that we are interested in?
> Graph partitioning operators for Gelly
> --------------------------------------
>
> Key: FLINK-1536
> URL: https://issues.apache.org/jira/browse/FLINK-1536
> Project: Flink
> Issue Type: New Feature
> Components: Gelly
> Reporter: Vasia Kalavri
> Assignee: Ivan Mushketyk
> Priority: Minor
>
> Smart graph partitioning can significantly improve the performance and
> scalability of graph analysis applications. Depending on the computation
> pattern, a graph partitioning algorithm divides the graph into (maybe
> overlapping) subgraphs, optimizing some objective. For example, if
> communication is performed across graph edges, one might want to minimize the
> edges that cross from one partition to another.
> The problem of graph partitioning is a well studied problem and several
> algorithms have been proposed in the literature. The goal of this project
> would be to choose a few existing partitioning techniques and implement the
> corresponding graph partitioning operators for Gelly.
> Some related literature can be found [here|
> http://www.citeulike.org/user/vasiakalavri/tag/graph-partitioning].
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
