[CSC 435] CSC 435 MPI Question

[Andrew J. Pounds]

And remember -- PBS is just executing a bash script -- there's nothing
preventing you from a using a couple of nested loops around your mpirun
command to set the number of threads and the matrix size.  It may be
kinda late for that now -- but that's a trick I've used before in these
type of scenarios.


On 4/21/20 8:18 PM, Andrew J. Pounds wrote:
> I don't think that is going to work -- but you can try it.   Cleanup
> could be ugly though.
>
> If you want to streamline your testing I don't recommend having a
> single executable call MPI_Initialize and MPI_Finalize repeatedly --
> you should generally only do that once.  You can, however, mimic what
> we did in the BabyBLAS and OpenBLAS assignments and have the size of
> the matrix and
> the number of OpenMP threads to used per processor be COMMAND LINE
> ARGUMENTS.    That should be a minimal code addition and you can copy
> the relevant portions from the fortran codes we used before.
>
> You could then have a PBS script for executing on 1 processor, another
> for 2 processors, another for 3 processors, etc.,  and inside that
> scripts call the same program over and over again --  but use a
> different matrix size and number of OpenMP threads per processor in
> each call. 
>
> I expect you to really hit these processors hard, so I made sure that
> before I left 218 today that the AC was on its lowest setting!
>
>
> During our time tomorrow afternoon I hope to show you how I had to
> structure a program to "re-use" the processes I had created on each node.
>
>
> On 4/21/20 7:14 PM, William Carl Baglivio wrote:
>> Dr. Pounds,
>>
>> One of the factors of MPI that I've been trying to stay mindful of is
>> making sure that the MPI stuff is fully closed out (so I don't leave
>> any fragments behind). In the main section of mmm_mpi.c, I want to
>> run the tests over a range of dimensions as opposed to having to go
>> back and define the dimension every time. I am good to loop the code
>> after the MPI_Finalize() call? I.e.
>>
>> For (int I = 2000; I < 10000; I += 2000)
>> {
>>      setting up code; running mmm; returning results; MPI_Finalize();
>> }
>>
>> ~Will B.
>
>
> -- 
> 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


-- 
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/cdb3debe/attachment.html>