[
https://issues.apache.org/jira/browse/MAPREDUCE-4049?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13447581#comment-13447581
]
Avner BenHanoch commented on MAPREDUCE-4049:
--------------------------------------------
_Asokan,_
My design has no conflict with your design. Below is a comment *I wrote you 4
months ago*:
{quote}
_Your patch for the trunk is good enough for my needs. I can write my RDMA
shuffle plugin based on either your patch or based on my patch. Hence, I am not
planning to submit additional patch for the trunk on top of your patch. (I will
only submit patch for 1.x)_
{quote}
(I have only now submitted a patch for the trunk, because of Arun's/Todd's
comment on my 1.x patch)
The academic paper I pointed as *Reference* should not be confused with my
plugin (Personally, I consider code in academic researches as POC and not as
product). The two relevant conclusions I take from this academic research are:
1) Hadoop can benefit from RDMA shuffle and shuffle plugin-ability
2) With fast shuffle, Hadoop can benefit from *additional* merge algorithms
that are not practical with slow shuffle.
That's all! There is no request for Hadoop to keep its coupling of shuffle
with merge.
Again, I encourage your decoupling! When your patch will be accepted to the
trunk, I will adjust future versions of my plugin following your decoupling.
*My design should not disturb you in any way!*
When reviewing my design from ReduceTask.java point of view, *If you merely
rename: ShuffleConsumerPlugin -> ReduceFeederPlugin, then you could easily see
that your decoupling design can peacefully come after my design.*
I believe the thing that disturbs you is that currently Hadoop uses 'shuffle'
which invokes 'merge' while you want the opposite direction. However, this is
outside the scope of my patch. Hence, you are welcome to build your patch on
top of mine. It is not really different than building your patch on top of the
current trunk.
I will be more than happy to assist you with anything you might need, and I'll
appreciate it if you gave me your blessing for my commit :-)
> plugin for generic shuffle service
> ----------------------------------
>
> Key: MAPREDUCE-4049
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-4049
> Project: Hadoop Map/Reduce
> Issue Type: Improvement
> Components: performance, task, tasktracker
> Affects Versions: 1.0.3, 1.1.0, 2.0.0-alpha, 3.0.0
> Reporter: Avner BenHanoch
> Labels: merge, plugin, rdma, shuffle
> Attachments: HADOOP-1.x.y.patch, Hadoop Shuffle Consumer Plugin
> TLD.rtf, Hadoop Shuffle Provider Plugin TLD.rtf, mapred-site.xml,
> mapreduce-4049.patch, mapreduce-4049.patch
>
>
> Support generic shuffle service as set of two plugins: ShuffleProvider &
> ShuffleConsumer.
> This will satisfy the following needs:
> # Better shuffle and merge performance. For example: we are working on
> shuffle plugin that performs shuffle over RDMA in fast networks (10gE, 40gE,
> or Infiniband) instead of using the current HTTP shuffle. Based on the fast
> RDMA shuffle, the plugin can also utilize a suitable merge approach during
> the intermediate merges. Hence, getting much better performance.
> # Satisfy MAPREDUCE-3060 - generic shuffle service for avoiding hidden
> dependency of NodeManager with a specific version of mapreduce shuffle
> (currently targeted to 0.24.0).
> References:
> # Hadoop Acceleration through Network Levitated Merging, by Prof. Weikuan Yu
> from Auburn University with others,
> [http://pasl.eng.auburn.edu/pubs/sc11-netlev.pdf]
> # I am attaching 2 documents with suggested Top Level Design for both plugins
> (currently, based on 1.0 branch)
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
