<html><head><style type='text/css'>p { margin: 0; }</style></head><body><div style='font-family: times new roman,new york,times,serif; font-size: 12pt; color: #000000'>Stefan,<br><br>I believe what you describe is referred to as performance variability. This is a known phenomenon and it is not specific to SCIP.<br>See for instance http://coral.ie.lehigh.edu/~jeff/mip-2008/talks/danna.pdf<br>See also Section 5 of Mixed Integer Programming Library version 5, Koch et al, Math. Prog. Comp. (2011).<br><br>I hope this helps.<br>Pierre<br><br><hr id="zwchr"><blockquote style="border-left:2px solid #1010FF;margin-left:5px;padding-left:5px;color:#000;font-weight:normal;font-style:normal;text-decoration:none;font-family:Helvetica,Arial,sans-serif;font-size:12pt;"><b>From: </b>"Stefan Lörwald" <stefan.loerwald@gmail.com><br><b>To: </b>"SCIP Mailingliste" <scip@zib.de><br><b>Sent: </b>Wednesday, November 20, 2013 11:11:29 AM<br><b>Subject: </b>[Scip] Inconsistent solving on different computers<br><br><div dir="ltr">Dear SCIP Developers,<br><br>a friend of mine recently discovered that SCIP behaves differently on different machines.<br><br>Although on every tested machine, the solution value is identical, the solution process differs quite radically. This are the results for the LOP example with problem ex1.mat<br>
<br>Virtual machine (64 bit)<br><br>Solutions found : 11 (11 improvements)<br>First Solution : +2.65100000000000e+03 (in run 1, after 1 nodes, 0.29 seconds, depth 592, found by <shiftandpropagate>)<br>
Solving Time (sec) : 21.39<br>
Solving Nodes : 506<br>Root Dual Bound : +3.10268636989738e+03<br><br><br>Virtual machine (32 bit)<br><br>Solutions found : 10 (9 improvements(null))<br>First Solution : +2.65100000000000e+03 (in run 1, after 1 nodes, 0.38 seconds, depth 592, found by <shiftandpropagate>)<br>
Solving Time (sec) : 28.52<br>Solving Nodes : 546<br>Root Dual Bound : +3.10271332082478e+03<br><br>netbook (32 bit)<br><br>Solutions found : 11 (10 improvements(null))<br>First Solution : +2.65100000000000e+03 (in run 1, after 1 nodes, 3.17 seconds, depth 592, found by <shiftandpropagate>)<br>
Solving Time (sec) : 245.61<br>Solving Nodes : 1093<br>Root Dual Bound : +3.10271332082478e+03<div><br></div><div>Our expectation was that both the number of solutions and the number of solving nodes should be identical, but they aren't.</div>
<div>The actual solving time is of course irrelevant because it depends heavily on the hardware.</div><div>To my knowledge, on every platform the code was compiled with the newest version of SCIP / SoPlex available.</div>
<div><br></div><div>Now two possible causes come to my mind:</div><div><br></div><div>Undefined behaviour and/or unspecified behaviour.</div><div><br></div><div>So the question is: Is this behaviour intended? If so, why? For comparability it seems logical to me to compare the number of solving nodes rather than solving time. That's because time measurements are not portable (practically always scaled). However the number of logical steps the algorithm takes (if deterministic) should be the same.</div>
<div>Asked provocatively: How can I ever make statements about the performance if it's not reproducible (at least up to scaling)?</div><div><br></div><div><br></div><div>Thanks in advance for clarification,</div><div>
Stefan</div></div>
<br>_______________________________________________<br>Scip mailing list<br>Scip@zib.de<br>http://listserv.zib.de/mailman/listinfo/scip<br></blockquote><br></div></body></html>