[
https://issues.apache.org/jira/browse/HBASE-8836?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13772444#comment-13772444
]
Tianying Chang commented on HBASE-8836:
---------------------------------------
Hi, Nicolas, my patch is trying to improve the write latency/throughput when
there are heavy read. My current understaning is read takes much longer than
write. (Note, I don't know the exact latency difference, but my understanding
is that write should be a lot faster than read. If this assumption is wrong,
then this patch has not value) My goal is the write latency/throughput should
be almost same when either there are 10,000 write-requests/sec or 5000
read-request + 5000 write-request. The short job should not wait behind a long
job if possible.
When there are heavy read load. E.g. more than 50% of all requests in the queue
are for read, then the write latency/throughput could be badly impact because a
thread who is handling the read request won't return back as fast compared with
if the thread is handling a write request. Separating the read and write into
two queues can make sure short job does not wait behind long job.
I will test the performance as WAL turned off as you suggested. I think that
could give us some more insight.
> Separate reader and writer thread pool in RegionServer, so that write
> throughput will not be impacted when the read load is very high
> -------------------------------------------------------------------------------------------------------------------------------------
>
> Key: HBASE-8836
> URL: https://issues.apache.org/jira/browse/HBASE-8836
> Project: HBase
> Issue Type: New Feature
> Components: Performance, regionserver
> Affects Versions: 0.94.8
> Reporter: Tianying Chang
> Assignee: Tianying Chang
> Fix For: 0.98.0
>
> Attachments: hbase-8836.patch, Hbase-8836-perfNumber.pdf
>
>
> We found that when the read load on a specific RS is high, the write
> throughput also get impacted dramatically, and even cause write data loss
> sometimes. We want to prioritize the write by putting them in a separate
> queue from the read request, so that slower read will not make fast write
> wait nu-necessarily long.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
