hama-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Edward J. Yoon" <edwardy...@apache.org>
Subject Re: New approach for mat-mat mult using mapreduce iterations
Date Mon, 06 Apr 2009 06:31:28 GMT
I expected that SIMDMultiplyMap/Reduce algorithm could be performed
sparse mat-mat mult well and efficiently through RPC over nodes but I
realized that's really bad algorithm for Hadoop/MapReduce.

I'm not sure exactly what kind of scheduling you are talking about,
but I guess the iterative method is best for sparse mat-mat mult.

On Sun, Mar 22, 2009 at 11:30 AM, Samuel Guo <guosijie@gmail.com> wrote:
> do you mean "one mr job one iteration"? I don't think it will improve
> the performance of mat-mat mult. the scheduling of the jobs will be a
> problem while multiplication.
>
> On 3/20/09, Edward J. Yoon <edwardyoon@apache.org> wrote:
>> Just considering, single table doesn't need much time to full scan.
>> So, if each row of matrix A have one mapreduce job, it'll be fast. (Of
>> course, there is a N time iterations.)
>>
>> for(int i = 0; i < A.getColumns(); i++) {
>>     // run Job
>> }
>>
>> What do you think?
>> --
>> Best Regards, Edward J. Yoon
>> edwardyoon@apache.org
>> http://blog.udanax.org
>>
>



-- 
Best Regards, Edward J. Yoon
edwardyoon@apache.org
http://blog.udanax.org

Mime
View raw message