You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@cassandra.apache.org by "Jonathan Ellis (Updated) (JIRA)" <ji...@apache.org> on 2011/10/07 04:50:29 UTC
[jira] [Updated] (CASSANDRA-3328) make leveled compaction suck less
for insert-heavy workloads
[ https://issues.apache.org/jira/browse/CASSANDRA-3328?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Jonathan Ellis updated CASSANDRA-3328:
--------------------------------------
Attachment: 3328.txt
Testing this idea shows that I was wrong: there really aren't very many non-overlapping compactions going on. In fact there is exactly one, the first L0 to L1 compaction. I guess the skipLevels code is already doing a good job of this for us.
> make leveled compaction suck less for insert-heavy workloads
> ------------------------------------------------------------
>
> Key: CASSANDRA-3328
> URL: https://issues.apache.org/jira/browse/CASSANDRA-3328
> Project: Cassandra
> Issue Type: Improvement
> Reporter: Jonathan Ellis
> Priority: Minor
> Attachments: 3328.txt
>
>
> In insert-heavy workloads, leveled compaction needs to do a lot of "promotion" from one level to a higher one, where the higher level has no overlapping sstables. We can optimize a lot of these away by not bothering to rewrite sstables that are younger than GCGraceSeconds.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira