You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@hive.apache.org by Tim Armstrong <ti...@gmail.com> on 2011/08/05 19:16:46 UTC
Review Request: Optimisation for RCFile reading to improve CPU usage.
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/1309/
-----------------------------------------------------------
Review request for hive and Ning Zhang.
Summary
-------
By tweaking the RCFile$Reader implementation to allow more efficient memory access I was able to reduce CPU usage. I measured the speed required to scan a gzipped RCFile, decompress and assemble into records. CPU time was reduced by about 7% for a full table scan, An improvement of about 2% was realised when a smaller subset of columns (3-5 out of tens) were selected.
This addresses bug HIVE-2350.
https://issues.apache.org/jira/browse/HIVE-2350
Diffs
-----
trunk/ql/src/java/org/apache/hadoop/hive/ql/io/RCFile.java 1154070
Diff: https://reviews.apache.org/r/1309/diff
Testing
-------
Ran TestRCFile unit test. Manually tested reading from warehouse table.
Thanks,
Tim
Re: Review Request: Optimisation for RCFile reading to improve CPU usage.
Posted by Tim Armstrong <ti...@gmail.com>.
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/1309/
-----------------------------------------------------------
(Updated 2011-08-11 23:07:05.774106)
Review request for hive, Yongqiang He and Ning Zhang.
Changes
-------
Minor change to avoid a compilation problem.
Summary
-------
By tweaking the RCFile$Reader implementation to allow more efficient memory access I was able to reduce CPU usage. I measured the speed required to scan a gzipped RCFile, decompress and assemble into records. CPU time was reduced by about 7% for a full table scan, An improvement of about 2% was realised when a smaller subset of columns (3-5 out of tens) were selected.
This addresses bug HIVE-2350.
https://issues.apache.org/jira/browse/HIVE-2350
Diffs (updated)
-----
trunk/ql/src/java/org/apache/hadoop/hive/ql/io/RCFile.java 1156839
Diff: https://reviews.apache.org/r/1309/diff
Testing
-------
Ran TestRCFile unit test. Manually tested reading from warehouse table.
Thanks,
Tim
Re: Review Request: Optimisation for RCFile reading to improve CPU usage.
Posted by Tim Armstrong <ti...@gmail.com>.
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/1309/
-----------------------------------------------------------
(Updated 2011-08-08 16:42:28.947782)
Review request for hive, Yongqiang He and Ning Zhang.
Summary
-------
By tweaking the RCFile$Reader implementation to allow more efficient memory access I was able to reduce CPU usage. I measured the speed required to scan a gzipped RCFile, decompress and assemble into records. CPU time was reduced by about 7% for a full table scan, An improvement of about 2% was realised when a smaller subset of columns (3-5 out of tens) were selected.
This addresses bug HIVE-2350.
https://issues.apache.org/jira/browse/HIVE-2350
Diffs
-----
trunk/ql/src/java/org/apache/hadoop/hive/ql/io/RCFile.java 1154070
Diff: https://reviews.apache.org/r/1309/diff
Testing
-------
Ran TestRCFile unit test. Manually tested reading from warehouse table.
Thanks,
Tim
Re: Review Request: Optimisation for RCFile reading to improve CPU usage.
Posted by Tim Armstrong <ti...@gmail.com>.
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/1309/
-----------------------------------------------------------
(Updated 2011-08-05 19:58:03.986031)
Review request for hive and Ning Zhang.
Changes
-------
Made some further optimisations to reduce branches and condition checks in the innermost loop, reducing CPU usage in getCurrentRow by about 20-25%, giving a few % more improvement overall for table scans.
Summary
-------
By tweaking the RCFile$Reader implementation to allow more efficient memory access I was able to reduce CPU usage. I measured the speed required to scan a gzipped RCFile, decompress and assemble into records. CPU time was reduced by about 7% for a full table scan, An improvement of about 2% was realised when a smaller subset of columns (3-5 out of tens) were selected.
This addresses bug HIVE-2350.
https://issues.apache.org/jira/browse/HIVE-2350
Diffs (updated)
-----
trunk/ql/src/java/org/apache/hadoop/hive/ql/io/RCFile.java 1154070
Diff: https://reviews.apache.org/r/1309/diff
Testing
-------
Ran TestRCFile unit test. Manually tested reading from warehouse table.
Thanks,
Tim