<html>
<head>
<meta http-equiv="Content-Type" content="text/html;
charset=windows-1252">
</head>
<body>
<div class="moz-cite-prefix">So there are ALL KINDS OF THINGS that
can go wrong if you play with the indexing of a hybrd MPI/OpenMP
program. If you are getting the correct trace of the matrix
(which should equal the dimension of the matrix) for dimensions of
like 4313 (and not just dimensions that are evenly divisible by
the number of threads) then you can generally say with confidence
that you are getting accurate results with your hybrid strategy.<br>
</div>
<div class="moz-cite-prefix"><br>
</div>
<div class="moz-cite-prefix">The problem that you may face is that
the techniques you take to guarantee that you are getting correct
results may impact your parallelism -- as there are regions to be
threaded in both the RANK=0 and RANK!=0 portions of the code. If
you simply demonstrate that your hybrid parallelism gives the
expected result and that you get improved performance when you to
move to more threads and nodes, then you have accomplished the
goal.</div>
<div class="moz-cite-prefix"><br>
</div>
<div class="moz-cite-prefix">There are MANY things that can hinder
this and I am not expecting any of you to achieve a linear
speedup. To do that we have would have to dive MUCH DEEPER into
inter-process locality, processor affinities, and processor
binding. Those are things I teach in the graduate version of this
course.</div>
<div class="moz-cite-prefix"><br>
</div>
<div class="moz-cite-prefix">I set the bar pretty low -- but it is
still a bar to get over. Can you achieve any speedup with correct
results when employing you hybrid parallel strategy?<br>
</div>
<div class="moz-cite-prefix"><br>
</div>
<div class="moz-cite-prefix">On 4/26/20 6:52 PM, wrote:<br>
</div>
<blockquote type="cite"
cite="mid:8eb00f4b225a44f2a90669e349545abc@BY5PR01MB5858.prod.exchangelabs.com">
<meta http-equiv="Content-Type" content="text/html;
charset=windows-1252">
<style type="text/css" style="display:none;"> P {margin-top:0;margin-bottom:0;} </style>
<div style="font-family: Calibri, Arial, Helvetica, sans-serif;
font-size: 12pt; color: rgb(0, 0, 0);">
Dr. Pounds,</div>
<div style="font-family: Calibri, Arial, Helvetica, sans-serif;
font-size: 12pt; color: rgb(0, 0, 0);">
<br>
</div>
<div style="font-family: Calibri, Arial, Helvetica, sans-serif;
font-size: 12pt; color: rgb(0, 0, 0);">
How should we demonstrate that we are getting accurate results?
Include the pbs resultant files in an appendix? It feels kind of
circular to say "let me demonstrate my results are accurate",
and then say "Ok I checked and they are." From my results I can
see that the accuracy is evident, but I don't know a good way to
demonstrate that to my audience. Thanks.</div>
<div style="font-family: Calibri, Arial, Helvetica, sans-serif;
font-size: 12pt; color: rgb(0, 0, 0);">
<br>
</div>
<div style="font-family: Calibri, Arial, Helvetica, sans-serif;
font-size: 12pt; color: rgb(0, 0, 0);"><br>
</div>
</blockquote>
<p><br>
</p>
<pre class="moz-signature" cols="72">--
Andrew J. Pounds, Ph.D. (<a class="moz-txt-link-abbreviated" href="mailto:pounds_aj@mercer.edu">pounds_aj@mercer.edu</a>)
Professor of Chemistry and Computer Science
Director of the Computational Science Program
Mercer University, Macon, GA 31207 (478) 301-5627
</pre>
</body>
</html>