[SATLUG] Re: Parallell Processing

[Steven Winston]

That depends on a lot of stuff.  If you're talking about using regular 
sequential programs in a multi-processing cluster.  Then you're looking at a 
system that must handle breaking up processes, send them to proper nodes, 
have those nodes handle them (memory could be shared a number of ways), 
without being told about the system it's being run on.  This means that it 
usually adds a lot of overhead to your programs.  The absolute best solution 
is to use programs written specifically for concurrency or parallel 
Processing.
Reason for this is that these programs are written to handle the above 
situations in a manner that delegates less overhead.  Remember when you're 
talking about this stuff, there's always a trade off.  It is possible to see 
perfect parallelism with distributed computing on a cluster (that's where you 
add up all your processers, and see the same results from your programs 
execution as what it would be on a sequential machine with the power of all 
processors together).  However, it's also possible to slow it down to slower 
than normal sequentialism.
On Wednesday 28 August 2002 11:09, Andrew C . Benson wrote:
> I was under the assumption the system would delegate certain processes to
> certain machines (based on utilization or something). And I believed that
> any process could be used, but only those built with multithreading in mind
> (pth, kth) would take special advantage of this.

-- 
Steven Winston
Home: (210) 641-0565
Office: (210) 582-5898
Cell: (281) 615-9612
http://www.cs.trinity.edu/~swinston/

 11:45am  up 27 days, 17:12,  2 users,  load average: 0.04, 0.10, 0.04

"The last time I make personal contact with a Minbari warship I sent
 it straight to Hell."
        -- Sheridan, "Points of Departure"