Github user wesolowskim commented on the issue:
https://github.com/apache/spark/pull/14137
Whole computation is iterative and depends on previous state of
sccWorkGraph. It iterates on sccWorkGraph.numVertices which is action under the
hood and without caching whole algorithm would be extremely slow, because with
every iteration sccWorkGraph would be computed all over again. Before I came
up with this solution I tried to unpersist intermediary sccWorkGraphs, and
colleague of mine tried to remove caches. Non of it worked, because of the
reasons mentioned above.
Initially i thought that caching should be removed but there are two
reasons that is not the case:
1. Scc algorithm requires knowledge of data (numVertices) that requires
action in spark
2. Every iteration is depended on previous state of graph
What is more pregel itself is implemented with caching and immediate
materialization.
If I cannot remove caches (I think I can't) additional ones have to be
added. I tested few other solutions and that is the one that turned to be the
best in terms of performance. I hope to find an optimal solution.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
