[
https://issues.apache.org/jira/browse/TINKERPOP-1290?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15271604#comment-15271604
]
Marko A. Rodriguez commented on TINKERPOP-1290:
-----------------------------------------------
Got {{GroupCountStep}} working with {{SparkStarBarrierInterceptor}}...
benchmark on the Blades over Friendster graph.
{code}
g.V().groupCount().by(outE().count())
before: 12 minutes
current: 2.6 minutes
{code}
> Create VertexProgramInterceptor as a pattern for GraphComputer strategies.
> --------------------------------------------------------------------------
>
> Key: TINKERPOP-1290
> URL: https://issues.apache.org/jira/browse/TINKERPOP-1290
> Project: TinkerPop
> Issue Type: Improvement
> Components: process
> Affects Versions: 3.2.0-incubating
> Reporter: Marko A. Rodriguez
>
> This is all being done in branch TINKERPOP-1288, but want to have a ticket
> for this feature for the release notes.
> A {{VertexProgramInterceptor}} can be used by a {{GraphComputer}} provider to
> NOT run a {{VertexProgram}} given some introspection into it. Instead, it can
> solve the {{VertexProgram}}'s problem using native methods instead of the
> overhead of the Gremlin OLAP {{GraphComputer}} semantics. The basic pattern
> is:
> {code}
> if(interceptor != null) {
> interceptor.apply(vertexProgram, graph, memory);
> } else {
> while(!vertexProgram.terminate(memory)) {
> vertexProgram.execute(...)
> }
> }
> {code}
> Thus far there is {{SparkStarBarrierInterceptor}} that can do Spark DSL
> computations for local star graph traversals that end with a
> {{ReducingBarrierStep}}.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
