You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@hbase.apache.org by "stack (JIRA)" <ji...@apache.org> on 2008/11/19 22:00:46 UTC
[jira] Created: (HBASE-1012) [performance] Try doctoring a
dfsclient so it shortcircuits hdfs when blocks are local
[performance] Try doctoring a dfsclient so it shortcircuits hdfs when blocks are local
--------------------------------------------------------------------------------------
Key: HBASE-1012
URL: https://issues.apache.org/jira/browse/HBASE-1012
Project: Hadoop HBase
Issue Type: Task
Reporter: stack
Ning Li up on list has stated that getting blocks using hdfs though the block is local takes almost the same amount of time as accesing the block over the network. See if can do something smarter when the data is known to be local short-circuiting hdfs if we can in a subclass of DFSClient (George Porter suggestion).
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Commented: (HBASE-1012) [performance] Try doctoring a
dfsclient so it shortcircuits hdfs when blocks are local
Posted by "Andrew Purtell (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-1012?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12791075#action_12791075 ]
Andrew Purtell commented on HBASE-1012:
---------------------------------------
Do we still want to do this?
> [performance] Try doctoring a dfsclient so it shortcircuits hdfs when blocks are local
> --------------------------------------------------------------------------------------
>
> Key: HBASE-1012
> URL: https://issues.apache.org/jira/browse/HBASE-1012
> Project: Hadoop HBase
> Issue Type: Task
> Reporter: stack
>
> Ning Li up on list has stated that getting blocks using hdfs though the block is local takes almost the same amount of time as accesing the block over the network. See if can do something smarter when the data is known to be local short-circuiting hdfs if we can in a subclass of DFSClient (George Porter suggestion).
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Commented: (HBASE-1012) [performance] Try doctoring a
dfsclient so it shortcircuits hdfs when blocks are local
Posted by "stack (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-1012?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12791122#action_12791122 ]
stack commented on HBASE-1012:
------------------------------
Leave this issue open I'd say. Todd Lipcon has made nice progress over in HADOOP-4801. There is critical pushback, "why such a difference -- maybe we should look more into why the canonical path is so slow", but the numbers are too good to leave on the table.
> [performance] Try doctoring a dfsclient so it shortcircuits hdfs when blocks are local
> --------------------------------------------------------------------------------------
>
> Key: HBASE-1012
> URL: https://issues.apache.org/jira/browse/HBASE-1012
> Project: Hadoop HBase
> Issue Type: Task
> Reporter: stack
>
> Ning Li up on list has stated that getting blocks using hdfs though the block is local takes almost the same amount of time as accesing the block over the network. See if can do something smarter when the data is known to be local short-circuiting hdfs if we can in a subclass of DFSClient (George Porter suggestion).
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.