Dramatic slowdown for -np 2

Jared Hodge jared_hodge at iat.utexas.edu
Tue Feb 12 10:28:01 EST 2002


You know your code a lot better than we possibly could, but it isn't
uncommon for there to be a drop in performance for 2 nodes (although
your case does seem extreme).  When you go to 2 nodes you have to pay
all the cost of parallelization, with the minimal benefits.  Not only do
none of the communications have to be executed in serial, but the
algorithm may actually be different (did you write the code or did
someone else?).  For instance, in a problem I was discussing with a
professor last week, the parallel version requires 3 times as much work
(to make it parallelizeable) so doing it on 2 processors would see a
performance hit, but it's worth it to do it on 100.  Even if you wrote
the code, it's possible that libraries that you may use (like Scalapack)
have separate parallel and serial versions.
In your case, I doubt this is the entire problem, since your
experiencing such a dramatic slowdown, but it could be part of it.  What
is the speedup for 3 nodes (If it's less than 1.5 this is probably the
entire problem)?  Best of luck finding the problem.

J Harrop wrote:
> 
> This may be an MPI problem, but I'm not sure so I'm posting it here and
> comp.parallel.mpi
> 
> We are developing an application on a four node Beowulf while we wait for
> the remaining nodes to arrive.  Speed-up has been close to predicted with
> -np 3 and 4 in a master/slave mode.  But when I run at -np 2 the speed
> drops to approximately 1/4 of the original serial application.  (On 4 nodes
> - that is 1 master and 3 slaves, we get about 2.5 times speed-up relative
> to the original application.)  All runs produce the same answer.
> 
> In the MPI code we have basic SEND, RECV, BCAST and REDUCE - nothing
> fancy.  Does anyone know if any of these or other MPI functions run into
> problems with one to one master slave ration?  Any other enlightenment
> would be welcome.
> 
> Cheers,
> 
> John Harrop
> 
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf

-- 
Jared Hodge
Institute for Advanced Technology
The University of Texas at Austin
3925 W. Braker Lane, Suite 400
Austin, Texas 78759

Phone: 512-232-4460
Fax: 512-471-9096
Email: Jared_Hodge at iat.utexas.edu
_______________________________________________
Beowulf mailing list, Beowulf at beowulf.org
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf



More information about the Beowulf mailing list