[
https://issues.apache.org/jira/browse/HADOOP-3053?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12580641#action_12580641
]
Dennis Kubes commented on HADOOP-3053:
--------------------------------------
The current patch is not ready for production. It is for first look, seeing
the overall architecture. Some benefits of this new framework are:
1) No blocking on reads, all reads and writes are handled in their own thread.
With current server reads are handled serially and writes in parallel.
2) Protocol is abstracted. Allows for protocol decoration if desired or for
completely new protocols. Also changing the protocol doesn't require an
understanding of the server internals.
3) High performance design. Not limited by a number of handlers. Processing
threads will block on queue insertion but can still accept connections.
Standard writes are queued then drained.
That is a basic list. The main idea is that this will be a cleaner, high
performance framework which will allow evolution of the RPC protocol.
Dennis
> New Server Framework for Hadoop RPC
> -----------------------------------
>
> Key: HADOOP-3053
> URL: https://issues.apache.org/jira/browse/HADOOP-3053
> Project: Hadoop Core
> Issue Type: Improvement
> Components: ipc
> Affects Versions: 0.17.0
> Environment: All
> Reporter: Dennis Kubes
> Assignee: Dennis Kubes
> Fix For: 0.17.0
>
> Attachments: HADOOP-3053-0-rough.patch
>
>
> This is a new Server framework for Hadoop RPC which replaces the current
> Server class. This new framework uses a reactor model to allow better
> throughput and better handling of clients.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
