[
https://issues.apache.org/jira/browse/HBASE-3855?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13029787#comment-13029787
]
dhruba borthakur commented on HBASE-3855:
-----------------------------------------
I have a cf with three columns. I insert 100K puts on a single rowkey with
random column values. all these kvs are cached in the memstore. A single scan
to retrieve the values for two specified columnname takes about 4 millisecons
of CPU without this patch and about 0.025 millieseconds of cpu with this patch.
Unfortunately, I cannot share the test program because it is intertwined with
quite a few non-opensource-able code.
However, I think the patch will benefit all cases because it is better to do a
O(logn) lookup rather then even a small number of sequential scans.
> Performance degradation of memstore because reseek is linear
> ------------------------------------------------------------
>
> Key: HBASE-3855
> URL: https://issues.apache.org/jira/browse/HBASE-3855
> Project: HBase
> Issue Type: Improvement
> Reporter: dhruba borthakur
> Assignee: dhruba borthakur
> Priority: Blocker
> Fix For: 0.94.0
>
> Attachments: memstoreReseek.txt
>
>
> The scanner use reseek to find the next row (or next column) as part of a
> scan. The reseek code iterates over a Set to position itself at the right
> place. If there are many thousands of kvs that need to be skipped over, then
> the time-cost is very high. In this case, a seek would be far lesser in cost
> than a reseek.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
