[
https://issues.apache.org/jira/browse/FLINK-3447?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15262431#comment-15262431
]
Vasia Kalavri commented on FLINK-3447:
--------------------------------------
For implementations whose performance highly depends on the underlying graph,
it makes sense to provide multiple implementations imo. For example, GSA
versions usually work better for skewed graphs, while scatter-gather / pregel
work better for graphs with non-skewed key distributions. For triangle
counting, I would only keep the most efficient one. It's true that we
refactored the gelly library and examples heavily in the past, but that was
while gelly was still in beta. It's still not part of the "stable" interfaces,
but we should be more careful now.
Bottom line is I'm not against organizing in subpackages, if you think it
simplifies things. BTW, what does 'asm' stand for?
> Package Gelly algorithms by framework
> -------------------------------------
>
> Key: FLINK-3447
> URL: https://issues.apache.org/jira/browse/FLINK-3447
> Project: Flink
> Issue Type: Improvement
> Components: Gelly
> Affects Versions: 1.0.0
> Reporter: Greg Hogan
> Assignee: Greg Hogan
> Priority: Minor
>
> Currently algorithms in the Gelly library are collected in the
> {{org.apache.flink.graph.library}} package. The gather-sum-apply class names
> are prefixed by "GSA". Gelly contains multiple frameworks as named in
> FLINK-3208.
> Since algorithms can be (and are) duplicated across the multiple frameworks,
> we can move the algorithms into subpackages by the name of the framework.
> - vertex-centric model: {{org.apache.flink.graph.library.pregel}}
> - scatter-gather model: {{org.apache.flink.graph.library.spargel}}
> - gather-sum-apply model: {{org.apache.flink.graph.library.gsa}}
> - native methods: {{org.apache.flink.graph.library.asm}}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
