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 2009/03/10 21:52:50 UTC
[jira] Created: (HBASE-1253) Remove unnecessary, expensive column
matcher lookup
Remove unnecessary, expensive column matcher lookup
---------------------------------------------------
Key: HBASE-1253
URL: https://issues.apache.org/jira/browse/HBASE-1253
Project: Hadoop HBase
Issue Type: Improvement
Reporter: stack
Fix For: 0.19.1
We do lots of matching on columns. We have different kinds of matchers dependent on how the scanner was configured. Down in HAS, we have a map of family name to its matchers. There is no need of the Map since column matching is done at the family/store context only. Removing should save on some unnecessary CPU (I saw up to 10% in simple seek+scan test -- so hopefully this comes down to 5 or 4%).
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Resolved: (HBASE-1253) Remove unnecessary, expensive column
matcher lookup
Posted by "stack (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-1253?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
stack resolved HBASE-1253.
--------------------------
Resolution: Fixed
Committed to branch. Thanks for the review Evgeny.
> Remove unnecessary, expensive column matcher lookup
> ---------------------------------------------------
>
> Key: HBASE-1253
> URL: https://issues.apache.org/jira/browse/HBASE-1253
> Project: Hadoop HBase
> Issue Type: Improvement
> Reporter: stack
> Fix For: 0.19.1
>
> Attachments: 1253.patch
>
>
> We do lots of matching on columns. We have different kinds of matchers dependent on how the scanner was configured. Down in HAS, we have a map of family name to its matchers. There is no need of the Map since column matching is done at the family/store context only. Removing should save on some unnecessary CPU (I saw up to 10% in simple seek+scan test -- so hopefully this comes down to 5 or 4%).
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Commented: (HBASE-1253) Remove unnecessary, expensive column
matcher lookup
Posted by "Evgeny Ryabitskiy (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-1253?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12680709#action_12680709 ]
Evgeny Ryabitskiy commented on HBASE-1253:
------------------------------------------
I think everything is ok. This AbstaractScaner used only for one family, so there no need to have Map of Vectors that will contain only one Vector.
You can also try to rollback changes and monitor in debug value of okCols.size(). It is always 1.
> Remove unnecessary, expensive column matcher lookup
> ---------------------------------------------------
>
> Key: HBASE-1253
> URL: https://issues.apache.org/jira/browse/HBASE-1253
> Project: Hadoop HBase
> Issue Type: Improvement
> Reporter: stack
> Fix For: 0.19.1
>
> Attachments: 1253.patch
>
>
> We do lots of matching on columns. We have different kinds of matchers dependent on how the scanner was configured. Down in HAS, we have a map of family name to its matchers. There is no need of the Map since column matching is done at the family/store context only. Removing should save on some unnecessary CPU (I saw up to 10% in simple seek+scan test -- so hopefully this comes down to 5 or 4%).
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Updated: (HBASE-1253) Remove unnecessary, expensive column
matcher lookup
Posted by "stack (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-1253?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
stack updated HBASE-1253:
-------------------------
Attachment: 1253.patch
Purge map. Just use Vector of matchers (Vector should probably be purged since Vector is synchronized and I don't think it needs be).
> Remove unnecessary, expensive column matcher lookup
> ---------------------------------------------------
>
> Key: HBASE-1253
> URL: https://issues.apache.org/jira/browse/HBASE-1253
> Project: Hadoop HBase
> Issue Type: Improvement
> Reporter: stack
> Fix For: 0.19.1
>
> Attachments: 1253.patch
>
>
> We do lots of matching on columns. We have different kinds of matchers dependent on how the scanner was configured. Down in HAS, we have a map of family name to its matchers. There is no need of the Map since column matching is done at the family/store context only. Removing should save on some unnecessary CPU (I saw up to 10% in simple seek+scan test -- so hopefully this comes down to 5 or 4%).
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Commented: (HBASE-1253) Remove unnecessary, expensive column
matcher lookup
Posted by "Evgeny Ryabitskiy (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-1253?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12680691#action_12680691 ]
Evgeny Ryabitskiy commented on HBASE-1253:
------------------------------------------
Comment here have to be fixed according changes:
// Holds matchers for each column family. Its keyed by the byte [] hashcode
// which you can get by calling Bytes.mapKey.
private Vector<ColumnMatcher> matchers = new Vector<ColumnMatcher>();
> Remove unnecessary, expensive column matcher lookup
> ---------------------------------------------------
>
> Key: HBASE-1253
> URL: https://issues.apache.org/jira/browse/HBASE-1253
> Project: Hadoop HBase
> Issue Type: Improvement
> Reporter: stack
> Fix For: 0.19.1
>
> Attachments: 1253.patch
>
>
> We do lots of matching on columns. We have different kinds of matchers dependent on how the scanner was configured. Down in HAS, we have a map of family name to its matchers. There is no need of the Map since column matching is done at the family/store context only. Removing should save on some unnecessary CPU (I saw up to 10% in simple seek+scan test -- so hopefully this comes down to 5 or 4%).
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.