[
https://issues.apache.org/jira/browse/HBASE-7884?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13585193#comment-13585193
]
Nicolas Spiegelberg commented on HBASE-7884:
--------------------------------------------
clockfly: thanks for posting the perf numbers! Those numbers look promising
and should definitely be used if correct. I "think" I added that perf test as
a rough estimate. Mostly to check out the gains from adding combinatorial
hashing. Point being, I'm not sure how useful it will be as a microbenchmark
for a number of reasons:
http://stackoverflow.com/questions/504103/how-do-i-write-a-correct-micro-benchmark-in-java/
I'm a little out of date on trunk functionality (my company's on an older
branch), but I know the older branch doesn't have proper microbenchmarking
utilities. I'm sure it would be involved to add a proper microbenchmark suite.
An easier solution to test that JVM hotspot warmup and other factors aren't
influencing this microbenchmark is to just major compact a large file (>1GB)
and compare. This is how I did the original benchmarking. You can either use
LoadTestTool or CreateRandomStoreFile to create a large load and just issue a
MAJOR_COMPACT on a region. You should be able to forcibly major compact a
single file over and over again for control. @Ted any idea on an easier
solution.
> ByteBloomFilter's performance can be improved by avoiding multiplication when
> generating hash
> ----------------------------------------------------------------------------------------------
>
> Key: HBASE-7884
> URL: https://issues.apache.org/jira/browse/HBASE-7884
> Project: HBase
> Issue Type: Bug
> Components: Performance
> Affects Versions: 0.94.5
> Reporter: clockfly
> Priority: Minor
> Fix For: 0.96.0, 0.94.6
>
> Attachments: bloom_optimization_trunk_patch.patch,
> bloom_performance_tunning.patch, hbase-7884-performance-report.pdf
>
>
> ByteBloomFilter's performance can be optimized by avoiding multiplication
> operation when generating hash
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
