[GAP Forum] StructureDescription & memory

R. Keith Dennis dennis at rkd.math.cornell.edu
Tue Apr 8 14:42:30 BST 2008

Over the past year I've been doing a large number of computations
with GAP and as I require much of the same data to be computed
over & over again, I've been storing it.  One of the items I've
been storing is the StructureDescription of a large number of
groups.  I realize that for groups, especially p-groups, the 
information obtained isn't always that useful, but sometimes it is.
For more complicted groups (lots of factors in the order) this
seems to be very expensive to compute, and very slow (hence the
reason to store it).

I've run into a problem, which may be intrinsic to the problem, may
be a problem with gap, or might be a problem with our computer or
operating system.  Perhaps someone can tell me which.

Ok, a perhaps stupid computation, but let's ignore that part:  
I've computed the structure descriptions of all the groups of order
256 (56092 total) except for the ones numbered 56083 through 56087.  I
believe all of these are of rank 7.  I've now either run out of memory
and had GAP quit, or I've brought the machine down 3 times when trying
to compute the result for 56083 or 56087.  With memory set at a max of
50G the computation runs out of memory and quits.  I've tried 100G and
80G with the same result - a crash.  The machine does have 128G of ram
and several times that in swap space (as well as 16 CPUs).

Any ideas what the problem may be?

One further question and a remark:

Are there any other implementations of a computation of structure
description that gives more useful information that that in gap?

I still believe that it makes sense to have a community accessible
archive of many such "standard" computations as structure
descricption, although when I raised that question at the recent
computational group theory conference in Ohio, there was not a great
deal of enthusiasm for the idea.  With the current cheap price of
storage (especially compared to that of memory and the time expended
for comutation), it would seem to make sense, to me at least.


More information about the Forum mailing list