[
https://issues.apache.org/jira/browse/HADOOP-7557?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13100506#comment-13100506
]
Todd Lipcon commented on HADOOP-7557:
-------------------------------------
Yea, I guess PBs also have type information. But type information means the
value can be much shorter as well.
bq. I also question whether the size of RPC headers is a significant factor in
Hadoop performance. Even when the payload is only a few bytes, the dominant
cost is not network bandwidth nor request parsing.
In Hadoop job performance, I agree. In something like HBase, though, it is a
significant amount of the time spent when requests are doing small reads out of
cache. Currently HBase has a fork of Hadoop IPC, but it would be good to keep
that use case in mind in hopes that down the road, we can converge
implementations again.
It seems the easiest way to settle the performance debate is to do a benchmark?
> Make IPC and RPC headers be extensible
> ---------------------------------------
>
> Key: HADOOP-7557
> URL: https://issues.apache.org/jira/browse/HADOOP-7557
> Project: Hadoop Common
> Issue Type: Improvement
> Reporter: Sanjay Radia
>
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
