[CSC 435] Lecture Postponed

[Andrew J. Pounds]

On 4/20/20 11:58 PM,  wrote:
> I was going to ask this question during class, but I'll go ahead and
> ask it now since I plan to get some good work done Tuesday. Once we
> find the optimal number of nodes to use, do we just need to then
> determine the optimal number of threads? Or is it one of those things
> where one affects the other so you still have to test? Thanks, see you
> Wednesday.
>
>

Yes -- this is one of those things that you have to test.  Essentially
you are trying to minimize a function with the matrix size being a
parameter and the number or systems and number of processors per system
being arguments.  The results depend on type of processor, cache layout,
memory hierarchy, network type, and network speed.

I mean think about it, if you had a slow network and really fast
processors, then you may only want to use a few systems.  If you had
something like a myrinet or infiniband based network then you may want
to push the data out to more systems and use fewer threads on each system.


You will all need to run numerous tests -- so this is not something to
wait on. 


I will have to make a short trip to campus today and I will reset the
system that fell off the network in lab 218.


-- 
Andrew J. Pounds, Ph.D.  (pounds_aj at mercer.edu)
Professor of Chemistry and Computer Science
Director of the Computational Science Program
Mercer University,  Macon, GA 31207   (478) 301-5627

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://theochem.mercer.edu/pipermail/csc435/attachments/20200421/d67d8d25/attachment.html>