You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@hbase.apache.org by "Jonathan Gray (JIRA)" <ji...@apache.org> on 2010/09/08 22:52:32 UTC
[jira] Created: (HBASE-2970) Meta blocks should be given their own
priority bucket in block cache
Meta blocks should be given their own priority bucket in block cache
--------------------------------------------------------------------
Key: HBASE-2970
URL: https://issues.apache.org/jira/browse/HBASE-2970
Project: HBase
Issue Type: Improvement
Components: io, regionserver
Reporter: Jonathan Gray
Assignee: Jonathan Gray
Fix For: 0.90.0
Bloom filters are currently using the LRU block cache. They are being treated as normal data blocks, however, it would be advantageous if we could treat me separately and give them their own portion of the LRU.
What we've seen in practice is that some blooms get very big and are accessed frequently. This leads to blooms consuming a majority of the cache so lots of churn on data blocks.
In any case, I think it is desirable in general to be able to control how much memory is used for blooms and other meta data.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Updated: (HBASE-2970) Meta blocks should be given their own
priority bucket in block cache
Posted by "stack (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-2970?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
stack updated HBASE-2970:
-------------------------
Fix Version/s: (was: 0.90.0)
Moving out of 0.90.
> Meta blocks should be given their own priority bucket in block cache
> --------------------------------------------------------------------
>
> Key: HBASE-2970
> URL: https://issues.apache.org/jira/browse/HBASE-2970
> Project: HBase
> Issue Type: Improvement
> Components: io, regionserver
> Reporter: Jonathan Gray
> Assignee: Jonathan Gray
>
> Bloom filters are currently using the LRU block cache. They are being treated as normal data blocks, however, it would be advantageous if we could treat me separately and give them their own portion of the LRU.
> What we've seen in practice is that some blooms get very big and are accessed frequently. This leads to blooms consuming a majority of the cache so lots of churn on data blocks.
> In any case, I think it is desirable in general to be able to control how much memory is used for blooms and other meta data.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.