[
https://issues.apache.org/jira/browse/HBASE-16417?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15904884#comment-15904884
]
ramkrishna.s.vasudevan commented on HBASE-16417:
------------------------------------------------
I was about read this doc but every time got distracted.
First question is that - just with default memstore in place we saw that Async
WAL was performing lesser than Sync WAL. Are you not seeing that? We saw that
behaviour with PE tool but till now I have not tested with YCSB. May be in
throughput we don't see a hit. That is good to see.
Next is that for the IHOP - 60 I believe for 40% memstore and 40% block cache
case (total 80%) this would have been less when you had both MSLAB and block
cache. Because you have already reserved 80% of your heap for these but still
you ask the GC to get triggered if the occupancy is 60%. So immediately we will
start getting GC and I think you will have mixed GCs.
But when MSLAB is not enabled then I believe IHOP - 60 is fine. I need to
verify this. Regarding GC pattern I have seen differences when the Xms is not
equal to Xmx. But that is another story.
Eager reducing number of WALs is something we need to explore. This YCSB does
not check for correctness of data so may be with AsycnWAL we need to run
LoadTestTool which has the option of checking the correctness.
And thanks for a great detailed report. Nice work.
> 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
>
>
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
