[
https://issues.apache.org/jira/browse/TINKERPOP-1287?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15565503#comment-15565503
]
Ted Wilmes commented on TINKERPOP-1287:
---------------------------------------
I have done further comparison between master and a stream free {{StarGraph}}
and determined there is no appreciable difference in runtime. The
{{String.intern()}} calls are showing up and since we have removed all of the
other uses of {{intern}} I think it's still worth taking these out and I'll
submit a PR to do so. We could take the {{stream}} calls out even though I
can't demonstrate an effect on runtime performance, but I'm very hesitant to do
that at this point.
I have attached profiling information gathered from running a
{{g.V().out().out().count()}} on {{SparkGraphComputer}} (single node) against a
graph with 1 million vertices and 10 million edges. Large portions of time are
spent deserializing and then I need to do some further digging to confirm, but
I believe a not insignificant amount of time is being spent on
{{NoSuchElementException}} handling. I have seen similar overhead using
{{TinkerGraphComputer}}.
> StarGraph has an overdose of Stream usage.
> ------------------------------------------
>
> Key: TINKERPOP-1287
> URL: https://issues.apache.org/jira/browse/TINKERPOP-1287
> Project: TinkerPop
> Issue Type: Improvement
> Components: hadoop, structure
> Affects Versions: 3.2.0-incubating, 3.1.2-incubating
> Reporter: Marko A. Rodriguez
> Assignee: Ted Wilmes
> Fix For: 3.2.3
>
> Attachments: stage0.svg, stage1.svg, stage2.svg
>
>
> {{StarGraph}} is loaded with {{Stream}}-usage. Gutting streams from
> TinkerGraph made it much faster. It would be good if we did the same thing
> for {{StarGraph}}.
> This can go into tp31/ and upmerge to master/.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
