[
https://issues.apache.org/jira/browse/HBASE-12202?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14163844#comment-14163844
]
stack commented on HBASE-12202:
-------------------------------
bq. Already we raised a JIRA to make BRs to be used in HFileBlock. But not sure
if that is needed now.
Lets do some experiments, POC, to see what is best fit?
bq. Should we try for BR?
Lets draw out the pipeline. Can mark where BR is better fit and ditto BB (BR
does not preclude BB, right?)
bq. Also the thought is that we may not convert entire read path starting from
DFS client passed byte [] to be converted to ByteBuffer at the first stage.
Only when the block moves to L2 cache from there start serving the reads from
Offheap byte buffer. What do you think? Do you feel we can address from the
DFSClient API itself?
Not sure I follow [~ram_krish] Please say again.
Would be cool if we could get from dfsclient into DBBs, checksum, and then put
into BC. May necessitate a copy from the DFSClient DBB to our own BC DBB
because DFSClient will want its DBB back. Offheap copy of DBBs should be fast
though and no garbage? Anyways, just thinking out loud.
> Support DirectByteBuffer usage in HFileBlock
> --------------------------------------------
>
> Key: HBASE-12202
> URL: https://issues.apache.org/jira/browse/HBASE-12202
> Project: HBase
> Issue Type: Sub-task
> Components: regionserver, Scanners
> Reporter: Anoop Sam John
> Assignee: Anoop Sam John
> Fix For: 2.0.0, 0.99.1
>
>
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
