hama-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Minho Kim (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HAMA-975) Improvement of Async RPC
Date Tue, 08 Sep 2015 10:54:46 GMT

    [ https://issues.apache.org/jira/browse/HAMA-975?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14734613#comment-14734613
] 

Minho Kim commented on HAMA-975:
--------------------------------

That sounds good. As you know, it must be tested in comparison with current version.
I recommend that you measure the performance, such as throughput of network and execution
time, using examples in Hama.
For example, the example run by different input data size. The larger input data size, the
better advantage you can get to measure performance.

> Improvement of Async RPC
> ------------------------
>
>                 Key: HAMA-975
>                 URL: https://issues.apache.org/jira/browse/HAMA-975
>             Project: Hama
>          Issue Type: Improvement
>          Components: bsp core
>            Reporter: JongYoon Lim
>
> Hama has a feature of async IPC. 
> I found some points which have possibility to be improved as below.
> 1. Add netty encoder and decoder to lighten a load of the handler.
> 2. Consider using native transport, EpollEventLoopGroup instead of NioEventLoopGroup.

> 3. Using pooled buffer. 
> 4. Using ctx.* instead of channel.* 
> 5. Find and remove blocking codes. 
> 6. Async-fashioned RPC response 
> Also we can consider compress or json-style marshalling(unmarshalling). 
> But I'm not sure these changes always result in improvement of the performance.. so benchmark
should be provided to prove the improvement. 
> I'd like to divide this issue to small sub tasks. 
> WDYT? 



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message