On Mon, Aug 19, 2019 at 5:30 AM Daegyu Han <hdg9...@gmail.com> wrote:
> Hi all, > > To my understand, Hedged-Read is a technique for copying IO to a replica > when pread () is delayed to mitigate IO latency such as speculative > execution in MapReduce. > > The release note on the feature has a pretty good explanation https://issues.apache.org/jira/browse/HDFS-5776 > - Does Hedged-read really happen where it uses hbase databases? > - Does hedged-read often occur because of a bottleneck in HDFS or a > bottleneck in the storage itself? > - What is the best number to set the timeout and thread for Hedged-Read? > > On the above, yes, it happens; on #2, no good answer (varies); and on #3, start with defaults and monitor the associated metrics (mentioned in the above cited release notes). Thanks, S > Thank you > ᐧ >