[
https://issues.apache.org/jira/browse/HBASE-16417?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15692806#comment-15692806
]
Anoop Sam John commented on HBASE-16417:
----------------------------------------
1. Ya pls turn ON wal. The configs that was put before were copied from the
off heap write tests. We had some issues to solve there to use default WAL
along with offheap cells.. That is why turned OFF then. Now even our tests use
WAL write. Sorry for the mistake
2. Again some what related to WAL off.. The speed with which the writes were
happening was more. So 2 threads are not enough for flush. Ya when the
#flushThreads and/or #compactionThreads are more, the IO pressure will be more
and the garbage creation rate also.. We will suffer with more GCs then..
Suggestion would be turn ON WAL and play with diff #threads and see how it is..
Blocking store files changes was also in similar lines.
3. We use G1GC in tests. And workload is RW. Means the heap memory of RS will
be always above 80%. This will make more GCs to happen. The initial heap
occupancy factor what u have is 45% I guess.. 80% is too high. So my
suggestion was that HBase should think of making it down now. Or else we have
to say when using G1GC how to tune IHOP. A large value for it dilutes the need
of G1 itself. ie. predictable GC pause..
cc [[email protected]]
> 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
>
>
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
