[GAP Forum] How much memory is needed?

Alexander Hulpke hulpke at math.colostate.edu
Tue Oct 1 17:21:46 BST 2013



Dear Mark,

I have had problems where a computer with bigger RAM has helped, though frankly if I had not had that larger computer I probably would have been able to do with a somewhat smaller machine and splitting up the job. In the end the benefit of more memory is somewhat logarithmic, if you give 512GB a 10, then 128GB is a 9 or 9.5.

As far as ordering a computer, I think the question is of the tradeoff what you would have to forgo if buying the larger machine (at the funding level I have, 512GB would be too high a tradeoff; you are probably better funded) and whether you already have calculations that you know failed at 128GB.  However, frankly, again my experience with (say) coset enumeration is that if it did not terminate in Memory x the chance to terminate in memory 2x is somewhere 5% or less, so a factor of 4 in memory is really only buying you a better success chance of 20% or less on these kinds of calculations.

Particularly concerning the calculations I expect you will be doing, also be aware that GAPs built-in Coset enumeration, knuth-bendix and groebner bases are far less powerful than the standalone c-programs and that there are packages that provide interfaces to ACE, KBMAG and singular.

Also often calculations tend to run for a long time and our system administrator wanted to reboot machines for safety updates -- I am now using DMTCP (http://dmtcp.sourceforge.net) to be able to save a running jobs to disk and to restart it after the machine reboot.

Best,

    Alexander





More information about the Forum mailing list