[ 
https://issues.apache.org/jira/browse/CASSANDRA-792?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12833653#action_12833653
 ] 

Stu Hood commented on CASSANDRA-792:
------------------------------------

>From the patch:
----
The optional MergeFactor attribute is used to tune read vs write
performance for a ColumnFamily. The default value is 10. A lower
MergeFactor (minimum 2) will cause the ColumnFamily to be compacted
more frequently, leading to improved read performance at the cost of
decreased write performance. Accordingly, a higher MergeFactor (max 256)
will improve write performance, at the expense of read performance. For
example, a MergeFactor of 8 indicates that compaction should only happen
when more than 8 sstables of a similar size are present. MergeFactor
is directly tied to the number of file handles required by Cassandra: a
higher MergeFactor means more file handles.
----

Once #685 adds backpressure to storage proxy, this setting will have a much 
more significant effect on write performance, since writes will be 
backpressured by compactions.

> Configurable 'merge factor' for Column Families
> -----------------------------------------------
>
>                 Key: CASSANDRA-792
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-792
>             Project: Cassandra
>          Issue Type: New Feature
>            Reporter: Stu Hood
>             Fix For: 0.6
>
>         Attachments: 
> 0001-Use-logarithmic-bucket-sizes-and-always-compact-the-.patch, 
> 0002-Cap-memory-usage-for-all-SSTables-being-compacted.patch, 
> 0003-Hardcoded-per-cf-merge-factor.patch, 0004-Configurable-MergeFactor.patch
>
>
> Apache Lucene flushes and merges 'segments' in a manner that is quite similar 
> to the treatment of Cassandra's sstables.
> One interesting and powerful tunable that Lucene provides is 'mergeFactor':
> EDIT: better link: 
> http://hudson.zones.apache.org/hudson/job/Lucene-trunk/javadoc/core/org/apache/lucene/index/IndexWriter.html#setMergeFactor(int)

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

Reply via email to