[
https://issues.apache.org/jira/browse/HBASE-17083?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15663335#comment-15663335
]
Yu Li commented on HBASE-17083:
-------------------------------
The idea of hedge read is good but one thing to notice is the network usage.
Previously with 1Gb bandwidth and hedge read enabled in our production
environment we ever observed network congestion and disabled the feature
afterwards, and we never tried with 10Gb bandwidth yet.
Another thing to watch out is the hedge read thread pool size. IIRC, we at
least need to set the pool size to {{hbase.regionserver.handler.count}} or else
it might be a bottleneck point.
Hope these information helps for your testing sir [~stack], and hopefully we
could get some good default settings through carefully testing and assure good
document supplied to users in our ref guide.
> Enable hedged reads by default in hbase-2.0.0
> ---------------------------------------------
>
> Key: HBASE-17083
> URL: https://issues.apache.org/jira/browse/HBASE-17083
> Project: HBase
> Issue Type: Improvement
> Affects Versions: 2.0.0
> Reporter: stack
> Assignee: stack
> Priority: Critical
> Fix For: 2.0.0
>
>
> Hedged reads came in in hadoop-2.4. hbase-2.0.0 requires a later version so
> we can expect support to be preseent. This issue is about a bit of testing to
> show defined benefit when hedged reads are on and then enabling them by
> default in hbase-2.0.0.
> See http://hbase.apache.org/book.html#_hedged_reads for current state of the
> doc.
> This issue came up in conversation a few of us were having Thursday and then
> Lars George started asking me their state in the last 24 hours.
> Assigning myself to do the bit of testing and enabling.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
