Newsgroups: comp.parallel.pvm
From: Brett Close <close@eecs.ukans.edu>
Subject: Load Balancing on SGI Challenge-S
Date: Thu, 21 Mar 1996 09:46:07 -0600
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Message-ID: <315179BF.41C6@eecs.ukans.edu>

Hello-
This may very well be an idiotic question.  
But, hey...        consider the entertainment value!  

I am currently using PVM for parallel implementations of 
image processing.   In the distant future, the application will 
run on a heterogeneous cluster over ATM.  For various reasons, 
right now I'm doing a mock-up on an SGI-Challenge-S 
(6 200MHz processors).  I'm not _Mr_Parallel_Man_, so bear with me...

One of the major problems (of course) is load balancing.  
Here's the question(s) - it's sortof twofold:
1.  I cant seem to figure out a way to map A TASK to A PROCESSOR
    on the SGI Challenge-S.  It's not like on a CM-5 on which you
    must address specific processors.  Anyone know how I can do
    this on a Challenge-S ???  < read further before answering >
2.  Even if I could manage to designate a specific processor on the 
    Challenge-S, am I not correct in thinking that it has it's own
    method of hardware-load-balancing?  I would think that a shared-
    memory, multiprocessor machine would have built-in methodology
    for balancing workload across all of its resources.  And if that's
    the case, it seems that any approach at writing in load-balancing
    methodology into a PVM-application for operation on the     
    Challenge-S would be futile, redundant, and probably meet with 
    some abberrant results.  

ANY ASSISISTANCE IS GREATLY APPRECIATED !!!

Thanks very much, 
 -----------------------------------------------------------------------
| Brett H. Close        -- Graduate Teaching Assistant                  |
| 235 Snow Hall         -- Univ. of Kansas Dept. of EECS - Lawrence, KS |
| vox: (hm)913.842.4854 -- (wk)913.864.7382                             |
| net: close@eecs.ukans.edu                                             |
| web: http://www.eecs.ukans.edu/~close                                 |
 -----------------------------------------------------------------------

