[
https://issues.apache.org/jira/browse/DRILL-5899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16243050#comment-16243050
]
ASF GitHub Bot commented on DRILL-5899:
---------------------------------------
Github user ppadma commented on the issue:
https://github.com/apache/drill/pull/1015
@paul-rogers Thanks a lot for the review. Updated the PR with code review
comments. Please take a look.
Overall, good improvement with this change. Here are the numbers.
select count(*) from `/Users/ppenumarthy/MAPRTECH/padma/testdata` where
l_comment like '%a'
1.4 sec vs 7 sec
select count(*) from `/Users/ppenumarthy/MAPRTECH/padma/testdata` where
l_comment like '%a%'
6.5 sec vs 13.5 sec
select count(*) from `/Users/ppenumarthy/MAPRTECH/padma/testdata` where
l_comment like 'a%'
1.4 sec vs 5.8 sec
select count(*) from `/Users/ppenumarthy/MAPRTECH/padma/testdata` where
l_comment like 'a'
1.1.65 sec vs 5.8 sec
I think for "contains", improvement is not as much as others, probably
because of nested for loops. @sachouche changes on top of these changes can
improve further.
> Simple pattern matchers can work with DrillBuf directly
> -------------------------------------------------------
>
> Key: DRILL-5899
> URL: https://issues.apache.org/jira/browse/DRILL-5899
> Project: Apache Drill
> Issue Type: Improvement
> Components: Execution - Flow
> Reporter: Padma Penumarthy
> Assignee: Padma Penumarthy
> Priority: Critical
>
> For the 4 simple patterns we have i.e. startsWith, endsWith, contains and
> constant,, we do not need the overhead of charSequenceWrapper. We can work
> with DrillBuf directly. This will save us from doing isAscii check and UTF8
> decoding for each row.
> UTF-8 encoding ensures that no UTF-8 character is a prefix of any other valid
> character. So, instead of decoding varChar from each row we are processing,
> encode the patternString once during setup and do raw byte comparison.
> Instead of bounds checking and reading one byte at a time, we get the whole
> buffer in one shot and use that for comparison.
> This improved overall performance for filter operator by around 20%.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
