[
https://issues.apache.org/jira/browse/HBASE-10204?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13852364#comment-13852364
]
Andrew Purtell edited comment on HBASE-10204 at 12/19/13 12:13 AM:
-------------------------------------------------------------------
This part sounds like overlap: "or reading from dfsclient into DBBs which are
then put into an offheap block cache". And/or getting DBBs wrapped around
mlocked regions of HDFS blocks out of the new DataNode caching layer. Would be
great to hook up something available in the DFSClient, if that were possible,
as feeding an off heap block pool.
was (Author: apurtell):
This part sounds like overlap: "or reading from dfsclient into DBBs which are
then put into an offheap block cache". And/or getting DBBs wrapped around
mlocked regions of HDFS blocks out of the new DataNode caching layer.
> Offheap efforts
> ---------------
>
> Key: HBASE-10204
> URL: https://issues.apache.org/jira/browse/HBASE-10204
> Project: HBase
> Issue Type: Umbrella
> Reporter: stack
>
> This is an umbrella issue under which we tie all efforts at offheaping
> whether [~apurtell] and crews' effort at single off-heap allocation arena for
> blockcache and memstore through @nkeywal reading queries into direct
> bytebuffers or reading from dfsclient into DBBs which are then put into an
> offheap block cache.
> This issue should include links to discussion and design. It also should
> host targets such as that posited by [~vrodionov] in HBASE-10191 where going
> end-to-end offheap would imply a new read/write pipeline.
--
This message was sent by Atlassian JIRA
(v6.1.4#6159)
