[CSC 435] TIMING PROTOCOLS

[Andrew J. Pounds]

On your last paper many of you complained about the fact that the queues 
were so full that you got terrible timings. I suspect that what was 
really going on was that people were timing code at the same time others 
were doing development work. That problem is going to be exacerbated on 
this project because you are timing multiprocessor code.       We have 
an easy way to alleviate that problem that mimics what is done at 
national supercomputing centers.

1. DO your development work on COBRA.  Cobra should have the same 
compilers.  Assuming you have your ssh keys set up on cobra you should 
be able to just clone your repository over there, get everything like 
you want it and then push it back to gitlab and clone it on zeus.   The 
only difference is that the  atlas libraries that I tuned are in 
/pub/pounds/ATLAS/lib on cobra so you would have to make that small 
change to your makefile if you want to compile against them.

2.  People using the zeus processors for Bear Day should be done (they 
are still using the cluster, but not the zeus processors). This means 
that zeus should be cleared out for testing.

3. ONLY DO YOUR TIMINGS on ZEUS using the BATCH QUEUS and make sure to 
specify the number of processors your need in your  batch script.  If 
you need all 8 processors for up to two hours then you would have the 
following line in your PBS file

#PBS -l nodes=1:zeus:ppn=8,walltime=2:00:00

If you don't need all 8 processors then please don't request all of them 
as that blocks others from using the queues.  It is also imperative that 
you -- for the sake of your classmates -- never use more processors that 
you actually requested.  If you request 4 processors in the batch queue 
then don't run a program that uses more than 4 threads.

4. I recommend that you include your PBS job scripts in your repository 
-- that will make it very easy for you to go back and make modifications 
if needed.


-- 
Andrew J. Pounds, Ph.D.  (pounds_aj at mercer.edu)
Professor of Chemistry and Computer Science
Mercer University,  Macon, GA 31207   (478) 301-5627
http://faculty.mercer.edu/pounds_aj

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://theochem.mercer.edu/pipermail/csc435/attachments/20160407/1f700597/attachment.html>