You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@hbase.apache.org by "Tudor Scurtu (Updated) (JIRA)" <ji...@apache.org> on 2012/04/02 16:29:25 UTC

[jira] [Updated] (HBASE-5625) Avoid byte buffer allocations when reading a value from a Result object

     [ https://issues.apache.org/jira/browse/HBASE-5625?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Tudor Scurtu updated HBASE-5625:
--------------------------------

    Attachment: 5625v5.txt

@Zhihong:
Thanks for the review request. I actually had to make my own in order to upload the diff: https://reviews.apache.org/r/4607/

The performance actually depends on the system capabilities. It's hard to write a microbenchmark test for an issue that manifests itself on large I/O intensive jobs that put a lot of gc pressure. I implemented a few of Cosmin's suggestions.
                
> Avoid byte buffer allocations when reading a value from a Result object
> -----------------------------------------------------------------------
>
>                 Key: HBASE-5625
>                 URL: https://issues.apache.org/jira/browse/HBASE-5625
>             Project: HBase
>          Issue Type: Improvement
>          Components: client
>    Affects Versions: 0.92.1
>            Reporter: Tudor Scurtu
>            Assignee: Tudor Scurtu
>              Labels: patch
>         Attachments: 5625.txt, 5625v2.txt, 5625v3.txt, 5625v4.txt, 5625v5.txt
>
>
> When calling Result.getValue(), an extra dummy KeyValue and its associated underlying byte array are allocated, as well as a persistent buffer that will contain the returned value.
> These can be avoided by reusing a static array for the dummy object and by passing a ByteBuffer object as a value destination buffer to the read method.
> The current functionality is maintained, and we have added a separate method call stack that employs the described changes. I will provide more details with the patch.
> Running tests with a profiler, the reduction of read time seems to be of up to 40%.

--
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