[
https://issues.apache.org/jira/browse/HBASE-5915?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13266755#comment-13266755
]
Todd Lipcon commented on HBASE-5915:
------------------------------------
Before we commit this, we should run some actual workloads. There are two funky
things I noticed about the patch:
1) the test case always has the behavior that the first KV to be compared is in
cache while the second one is not, which is not quite realistic in a real
workload. Maybe it's better to time insertion of a few million KVs into a
KeyValueSkipListSet instead.
2) when I tried this patch with a YCSB workload, it actually took _more_ CPU
seconds than without the patch, which is odd given the microbenchmark. So
something isn't aligning with reality. I didn't have time to investigate that.
> Improve performance of KeyValue comparison
> ------------------------------------------
>
> Key: HBASE-5915
> URL: https://issues.apache.org/jira/browse/HBASE-5915
> Project: HBase
> Issue Type: Improvement
> Components: performance
> Affects Versions: 0.94.1
> Reporter: Todd Lipcon
> Assignee: Todd Lipcon
> Attachments: 5915-v2.txt, hbase-5915.txt
>
>
> Got bored tonight and poked around at making KeyValue comparison faster. I
> was able to get a ~23% speedup on a micro-benchmark I wrote by using Unsafe
> in a few more places. This should make some difference for real workloads,
> since this is usually one of the top CPU consumers.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
