[
https://issues.apache.org/jira/browse/HDFS-13053?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16338067#comment-16338067
]
Pulkit Misra commented on HDFS-13053:
-------------------------------------
Thanks for the pointer [~hanishakoneru]!
https://issues.apache.org/jira/browse/HDFS-10917 is different in a few ways
compared to what we are proposing.
The main difference is that in our approach the packet processing latency of
each datanode in the pipeline is propagated upstream to the client (and not the
namenode). This approach allows a client to determine whether a datanode in the
current pipeline is slow compared to others, getting slower over time etc. The
client can also take remedial action with such information and cut down write
tail latencies.
It would be great to have inputs from [~xiaobingo] and [~arpitagarwal] since
they worked on the related issues.
> Track time to process packet in Datanode
> ----------------------------------------
>
> Key: HDFS-13053
> URL: https://issues.apache.org/jira/browse/HDFS-13053
> Project: Hadoop HDFS
> Issue Type: Improvement
> Components: datanode
> Reporter: Íñigo Goiri
> Assignee: Pulkit Misra
> Priority: Minor
> Attachments: HDFS-13053.000.patch
>
>
> We should track the time that each datanode takes to process a packet.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
