[
https://issues.apache.org/jira/browse/HIVE-461?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12724461#action_12724461
]
Zheng Shao commented on HIVE-461:
---------------------------------
Some preliminary comments:
@hive-461-2009-06-26.patch:
1. Use ArrayList instead of Vector: the only difference is that Vector's
methods are synchronized. We don't need that
2. Use Integer.valueOf(int i) instead of new Integer(i) because that's more
efficient
3. RCFileInputFormat.setReadColumnIDs/setFullyReadColumns/getReadColumnIDs ->
Can you move these methods to some utility class? Conceptually this should not
be specific to RCFile.
4. setReadColumnIDs and getReadColumnIDs should use the same thing
(ArrayList<Integer> instead of int[])
5. TableScanOperator.java: Vector<Integer> neededColumnIDs; should be moved to
tableScanDesc.java (and use ArrayList<Integer>)
Have you figured out what's wrong with input20.q?
> Optimize RCFile reading by using column pruning results
> -------------------------------------------------------
>
> Key: HIVE-461
> URL: https://issues.apache.org/jira/browse/HIVE-461
> Project: Hadoop Hive
> Issue Type: Improvement
> Components: Serializers/Deserializers
> Affects Versions: 0.4.0
> Reporter: Zheng Shao
> Assignee: Yongqiang He
> Attachments: hive-461-2009-05-26.patch, hive-461-2009-06-26.patch
>
>
> RCFile is a column-based file format introduced in HIVE-352. Column-based
> storage has shown better compression ratio. On our internal data set (30
> columns, most of them are short integer strings), we are seeing
> gzip-compressed RCFile to be 20%+ smaller than gzip-compressed SequenceFile.
> RCFIle also has the potential to improve the reading efficiency a lot since
> it compresses each column separately.
> We should integrate RCFile with the column pruning results from Hive to make
> the reading faster.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.