[ https://issues.apache.org/jira/browse/HBASE-9969?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13825391#comment-13825391 ]
Chao Shi commented on HBASE-9969: --------------------------------- bq. I wonder if you should make ~10 cols/row, and NUM_KEYS_PER_SCANNER/10 rows. This should exacerbate the problem i mentioned above. I tried this on my laptop but seems your case above is even faster than before. Maybe there is something wrong with my environment. I will try it on my devbox tomorrow. bq. This literally sits at the core of HBase, and we'd better be a 100% sure it's OK. I'm totally agree with you! I think we can delay the optimization to future and make it correct first. As it is a so critical piece and the original implementation is so tricky (the bloomfilter part), I'd like to invite all folks to review this patch, particularity who know KeyValueHeap better than me (e.g. [~liyin]). > 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, 0.94.15 > > Attachments: 9969-0.94.txt, hbase-9969-v2.patch, hbase-9969-v3.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)