[Swift-devel] Falkon and Coaster support for MPI

[Ioan Raicu]

In the interest of productive arguing :)
...

But we have to prioritize our work!

I bet we could put a full time staff/student on surveying the state of 
the art scheduling strategies and to implement that in Falkon/Coaster, 
and it would likely take months (maybe more, if we find hurdles that our 
environment is posing to us that others didn't face due to a more 
general Linux environment) to do this task well. 

I am willing to spend days on finding a solution for this, but anything 
more, and I would have to say that its better to add MPI support to 
Coaster (regardless of what I really believe).  If the goal is some 
fancy scheduling algorithm (beyond the simple one I outlined), well I 
can save myself a few days of work, and give you/Ben the token to start 
working on MPI support in Coaster!

Ioan

Mihael Hategan wrote:
> On Mon, 2008-06-30 at 12:31 -0500, Ioan Raicu wrote:
>   
>> Comon Mihael, quit being so quick to argue :)\
>>     
>
> I'm not sure what you mean. Arguments are not a bad thing. They are the
> mechanism to clarify issues and solve problems. Arguing by exclusively
> using rhetoric or not following logic much, on the other hand, is
> probably not that useful.
>
> I was trying to point out that there are some recurring patterns when
> moving from the single site batch job pattern to a multi site grid-like
> pattern, and that there's little infrastructure to support that move.
> But the problems themselves are there to stay, and won't go away if
> ignored.
>
>   
>> Adding MPI support might be easy, or hard, but it will certainly add 
>> complexity to the scheduler that needs to support MPI.  If the user 
>> community asks for this feature, we (Swift, Falkon, etc) would 
>> eventually support it, but its not clear what priority this should take, 
>> which partly depends on how hard it is to achieve the desired goal 
>> (given all our limited resources and time).
>>
>> Ioan
>>
>> Mihael Hategan wrote:
>>     
>>>> I think we should be nervous about getting into the business of  
>>>> implementing scheduler functionality like this.
>>>>     
>>>>         
>>> That's funny. These are the problems we need to solve. We (Swift) are
>>> pretty much suffering from nobody else in the grid world having done so.
>>> I guess too many people were nervous.
>>>
>>>
>>>
>>>   
>>>       
>
>
>   

-- 
===================================================
Ioan Raicu
Ph.D. Candidate
===================================================
Distributed Systems Laboratory
Computer Science Department
University of Chicago
1100 E. 58th Street, Ryerson Hall
Chicago, IL 60637
===================================================
Email: iraicu at cs.uchicago.edu
Web:   http://www.cs.uchicago.edu/~iraicu
http://dev.globus.org/wiki/Incubator/Falkon
http://dsl-wiki.cs.uchicago.edu/index.php/Main_Page
===================================================
===================================================