[
https://issues.apache.org/jira/browse/HADOOP-1045?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12479937
]
Jim Kellerman commented on HADOOP-1045:
---------------------------------------
With respect to block caching, it is HStore that 'talks' to MapFile. It has no
notion of blocks as it just does a seek and a get. It would appear then that
what is needed is a MapFile.CachingReader, which in turn would require
SequenceFile.CachingReader, since MapFile.Reader merely tells the underlying
Sequence file to seek and return the record at the appropriate location. If a
SequenceFile.CachingReader could keep some (configurable) number of blocks in
memory, it would not have to keep getting them from DFS. This is safe as
SequenceFiles are immutable once they are closed after writing.
> Code for HBase
> --------------
>
> Key: HADOOP-1045
> URL: https://issues.apache.org/jira/browse/HADOOP-1045
> Project: Hadoop
> Issue Type: New Feature
> Environment: All environments
> Reporter: Mike Cafarella
> Assigned To: Jim Kellerman
> Attachments: hbase.patch, hbase.patch, hbase.patch, hbase.patch,
> hbase.tar.gz
>
>
> I've written some code for HBase, a BigTable-like file store. It's not
> perfect, but it's ready for other people to play with and examine.
> The attached tarball has the source and a README
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
