[
https://issues.apache.org/jira/browse/CASSANDRA-11550?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15448224#comment-15448224
]
Marcus Eriksson commented on CASSANDRA-11550:
---------------------------------------------
So, starting from 40gb unleveled data we get this with fanout = 6:
* 5% longer time to compact
* 10% more compaction bytes written
* 2% larger live size after compaction
I guess all those numbers are due to the fact that we end up with more levels
and we need to do more compaction to get the data into those levels
These numbers are not 100% fair either as what is most interesting is a
steady-state situation, but that is much harder to benchmark. I'll set up a 2
node cluster and change one of them to use fanout = 6 and run stress.
> Make the fanout size for LeveledCompactionStrategy to be configurable
> ---------------------------------------------------------------------
>
> Key: CASSANDRA-11550
> URL: https://issues.apache.org/jira/browse/CASSANDRA-11550
> Project: Cassandra
> Issue Type: New Feature
> Components: Compaction
> Reporter: Dikang Gu
> Assignee: Dikang Gu
> Labels: lcs
> Fix For: 3.x
>
> Attachments:
> 0001-make-fanout-size-for-leveledcompactionstrategy-to-be.patch
>
>
> Currently, the fanout size for LeveledCompactionStrategy is hard coded in the
> system (10). It would be useful to make the fanout size to be tunable, so
> that we can change it according to different use cases.
> Further more, we can change the size dynamically.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
