[
https://issues.apache.org/jira/browse/CASSANDRA-13664?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16076030#comment-16076030
]
Marcus Eriksson commented on CASSANDRA-13664:
---------------------------------------------
bq. isn't the issue here that the streams aren't weighted
Yes, that would be a nicer solution. It wasn't obvious to me how to do maximum
bipartite matching with weighted edges though so I went with the easy solution
(I guess having an edge for each token would be a way, but that would be quite
silly). Also have to say I didn't spend very much time trying to figure it out,
so if you have an idea, please let me know.
> RangeFetchMapCalculator should not try to optimise 'trivial' ranges
> -------------------------------------------------------------------
>
> Key: CASSANDRA-13664
> URL: https://issues.apache.org/jira/browse/CASSANDRA-13664
> Project: Cassandra
> Issue Type: Bug
> Reporter: Marcus Eriksson
> Assignee: Marcus Eriksson
> Fix For: 4.x
>
>
> RangeFetchMapCalculator (CASSANDRA-4650) tries to make the number of streams
> out of each node as even as possible.
> In a typical multi-dc ring the nodes in the dcs are setup using token + 1,
> creating many tiny ranges. If we only try to optimise over the number of
> streams, it is likely that the amount of data streamed out of each node is
> unbalanced.
> We should ignore those trivial ranges and only optimise the big ones, then
> share the tiny ones over the nodes.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
