[
https://issues.apache.org/jira/browse/HBASE-9969?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13823319#comment-13823319
]
Lars Hofhansl commented on HBASE-9969:
--------------------------------------
Tested again without performing a major compaction. In my scenario I have 10m
rows with 5 columns each. Ends up being only three regions and only that last
one has more than one store file (4 in this case).
With that I still did not see any improvement.
Re: HBASE-9778, it's still not immediately clear there how to optimize the
cases mentioned there (many small KVs) while keeping other optimizations (for
example for large KVs, where the likelihood is high that a call to next() will
land us unnecessarily in the next block).
Anything that will bring HBase's CPU consumption down is a win. Unless all data
is in the cache I would expect us always being CPU bound, that is not always
true (with SSDs for example).
> Improve KeyValueHeap using loser tree
> -------------------------------------
>
> Key: HBASE-9969
> URL: https://issues.apache.org/jira/browse/HBASE-9969
> Project: HBase
> Issue Type: Improvement
> Components: Performance, regionserver
> Reporter: Chao Shi
> Assignee: Chao Shi
> Fix For: 0.98.0, 0.96.1
>
> Attachments: hbase-9969-v2.patch, hbase-9969.patch, hbase-9969.patch,
> kvheap-benchmark.png, kvheap-benchmark.txt
>
>
> LoserTree is the better data structure than binary heap. It saves half of the
> comparisons on each next(), though the time complexity is on O(logN).
> Currently A scan or get will go through two KeyValueHeaps, one is merging KVs
> read from multiple HFiles in a single store, the other is merging results
> from multiple stores. This patch should improve the both cases whenever CPU
> is the bottleneck (e.g. scan with filter over cached blocks, HBASE-9811).
> All of the optimization work is done in KeyValueHeap and does not change its
> public interfaces. The new code looks more cleaner and simpler to understand.
--
This message was sent by Atlassian JIRA
(v6.1#6144)
