[
https://issues.apache.org/jira/browse/HBASE-16890?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15603895#comment-15603895
]
Duo Zhang commented on HBASE-16890:
-----------------------------------
In my opinion, there are two things may impact performance a lot. One is the
contention when append as [~ram_krish] said, and the second is the aggregation
of sync request as [~stack] said.
The former one is hard to optimize as we have some other logics in the critial
section, it is not only a queue offering...
The latter one is a trade off between latency and throughput. I have already
done the aggregation work in the consumer of AsyncFSWAL so first we need to
confirm that it works correctly... And second, if we add some delay before the
actual sync, it may help increasing the throughput. This is a common tech to
increase throughput by also increasing latency. Of course, it does not always
increase latency. Here sync is a heavy operation, so choosing a suitable delay
may also decrease the latency. I think we could have a try.
Thanks.
> Analyze the performance of AsyncWAL and fix the same
> ----------------------------------------------------
>
> Key: HBASE-16890
> URL: https://issues.apache.org/jira/browse/HBASE-16890
> Project: HBase
> Issue Type: Sub-task
> Components: wal
> Affects Versions: 2.0.0
> Reporter: ramkrishna.s.vasudevan
> Assignee: ramkrishna.s.vasudevan
> Fix For: 2.0.0
>
> Attachments: contention.png, contention_defaultWAL.png
>
>
> Tests reveal that AsyncWAL under load in single node cluster performs slower
> than the Default WAL. This task is to analyze and see if we could fix it.
> See some discussions in the tail of JIRA HBASE-15536.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
