<html><head><style type='text/css'>p { margin: 0; }</style></head><body><div style='font-family: Arial; font-size: 12pt; color: #000000'><div><br></div>----- Original Message -----<br>From: "Jeff Layton" <laytonjb@att.net><br>To: "Greg Lindahl" <lindahl@pbm.com><br><br>>><br>>> CFD codes come in many shapes and sizes, so generalizing about them is<br>>> not a good idea. Really.<br>>> <br>>I definitely agree. That's why I said "usually" :)<div><br></div><div>Sources of unpredictable (non-pipeline-able) latency in CFD</div><div>include turbulent flows that move areas requiring denser meshes</div><div>around your data partition space, chemistry and mixing that require</div><div>the conditional invocation different kernel physics, particle tracking,</div><div>and complicated fluid surface interactions among other things.</div><div><br></div><div>Of course, the latency we are talking about here is a multi-tiered</div><div>phenomenon involving non-node-local memory references, local</div><div>memory indexed or non-strided reference patterns, and the companion</div><div>cache-related complications.</div><div><br></div><div>rbw <br><br>_______________________________________________<br>Beowulf mailing list, Beowulf@beowulf.org sponsored by Penguin Computing<br>To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf<br></div></div></body></html>