[
https://issues.apache.org/jira/browse/HBASE-16417?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15929565#comment-15929565
]
Eshcar Hillel commented on HBASE-16417:
---------------------------------------
Updated benchmarks report.
You can see in Figure 8 that merging the indices of pipeline segments into a
single index (depicted by orange bars) eliminates the overhead of reading
multiple segments. In addition, figures 9 and 10 show that the performance of
basic and basic with merge are equivalent in write-only workload both for the
case of uniform distribution and for zipfian distribution.
I also ran write-only benchmarks with big values. As we anticipated the affect
of reducing the meta data decreases as the size of data itself increases. We
still see the same trend with respect to write amplification with eager policy
in zipfian distribution.
It means that the main benefit we see in performance in the experiment is from
reducing the gc. We almost don’t see any gain in throughput due to reduction in
compaction. This might be due to the hardware - SSD which handles disk
compaction well, or since we were not able to saturate the server enough so
that compaction becomes a problem. But in any case running with basic+merge or
eager is as good as running with no compaction, and we are not seeing any
degradation in performance.
> In-Memory MemStore Policy for Flattening and Compactions
> --------------------------------------------------------
>
> Key: HBASE-16417
> URL: https://issues.apache.org/jira/browse/HBASE-16417
> Project: HBase
> Issue Type: Sub-task
> Reporter: Anastasia Braginsky
> Assignee: Eshcar Hillel
> Fix For: 2.0.0
>
> Attachments: HBASE-16417-benchmarkresults-20161101.pdf,
> HBASE-16417-benchmarkresults-20161110.pdf,
> HBASE-16417-benchmarkresults-20161123.pdf,
> HBASE-16417-benchmarkresults-20161205.pdf,
> HBASE-16417-benchmarkresults-20170309.pdf,
> HBASE-16417-benchmarkresults-20170317.pdf
>
>
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
