[
https://issues.apache.org/jira/browse/HBASE-12090?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14150104#comment-14150104
]
Vladimir Rodionov commented on HBASE-12090:
-------------------------------------------
OK, here are some test results.
Two tables with one column family:
* Narrow Table: narrow row = 1 col + 1 version each
* Wide Table : wide row = 3 col + 10 versions each
Tests:
* Get - reads randomly 2M rows from a table fully cached in a block cache.
* Scan - makes full scan using RegionScannerImp on a table fully cached in a
block cache.
.h3 Test results - Patch vs. Trunk improvement (+%%)
|| Narrow table || Wide Table ||
| GET | +2% | +16% |
| SCAN | +14% | +26% |
.h3 Conclusions
Patched version is faster in all tests.
> Bytes: more Unsafe, more Faster
> --------------------------------
>
> Key: HBASE-12090
> URL: https://issues.apache.org/jira/browse/HBASE-12090
> Project: HBase
> Issue Type: Improvement
> Components: Performance
> Affects Versions: 0.94.23, 0.98.6
> Reporter: Vladimir Rodionov
> Assignee: Vladimir Rodionov
> Fix For: 2.0.0, 0.98.7, 0.94.24, 0.99.1
>
> Attachments: 12090-v1.1.txt, HBASE-12090.2.patch, HBASE-12090.patch
>
>
> Additional optimizations to *org.apache.hadoop.hbase.util.Bytes*:
> * New version of compareTo method.
> * New versions for primitive converters : putXXX/toXXX.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
