[
https://issues.apache.org/jira/browse/HDFS-6581?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14140845#comment-14140845
]
Colin Patrick McCabe commented on HDFS-6581:
--------------------------------------------
[~andrew.wang]: I agree that it would be nice to see a short-circuit write
benchmark. I don't think it's absolutely necessary to see that before merging,
though. But I do think some kind of benchmark of reading and writing from
in-memory replicas is necessary, since as you say, this is a
performance-focused project. At least __reading__ from these replicas should
be fast, similar to how reading from HDFS-4949 replicas is fast. Merging
without this information would be like merging the ACL branch without ever
trying to create an ACL. How do you know if it works?
I also think we absolutely need a better eviction strategy than LRU before
merging. Any kind of real-world workload will probably show this really
clearly.
Sorry for being a party pooper, but I wanted to bring up these concerns before
the merge vote. If this spends another week or two in development, it will be
a small price to pay for something we can be proud of.
> Write to single replica in memory
> ---------------------------------
>
> Key: HDFS-6581
> URL: https://issues.apache.org/jira/browse/HDFS-6581
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: datanode
> Reporter: Arpit Agarwal
> Assignee: Arpit Agarwal
> Attachments: HDFS-6581.merge.01.patch, HDFS-6581.merge.02.patch,
> HDFS-6581.merge.03.patch, HDFS-6581.merge.04.patch, HDFS-6581.merge.05.patch,
> HDFS-6581.merge.06.patch, HDFSWriteableReplicasInMemory.pdf,
> Test-Plan-for-HDFS-6581-Memory-Storage.pdf
>
>
> Per discussion with the community on HDFS-5851, we will implement writing to
> a single replica in DN memory via DataTransferProtocol.
> This avoids some of the issues with short-circuit writes, which we can
> revisit at a later time.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
