[MITgcm-support] pbs with exp0

[Ed Hill]

On Fri, 2006-07-07 at 20:17 -0400, Patrick Heimbach wrote:
> 
> a random guess: Can you type
> size mitgcmuv
> and send the numbers.


Hi Nicholas,

As Patrick mentions, its seems you're running out of memory (stack?
heap? hard to say...) and that the error isn't handled gracefully.

Just to see what would happen, I compiled tutorial_barotropic_gyre (the
new name for the old exp0) with g77 on an x86 machine and with sNx,sNy
both equal to 1000 and then ran it on a machine with 2GB RAM.  It
immediately seg-faults, even when the stack size is set to unlimited.
So, thats another compiler that produces code which doesn't gracefully
handle the too-large memory situation.  :-)

If you need to run a very large setup, then the idea is to split it into
smaller tiles and then run each on separate machines.  As Patrick
suggests, MPI isn't just a way to get more CPU cycles -- it also allows
big-memory problems to be made feasible.

Ed

ps - Have you tried unlimiting the stack?  In bash the command is:
     "ulimit -s unlimited" or in [t]csh its:
     "limit stacksize unlimited".

pps - If you search our email list Constantinos has posted notes 
      about how to specify the memory model (small, medium, huge) 
      which can be done with some compilers/architectures to use 
      larger pointers, etc. which might help you.

-- 
Edward H. Hill III, PhD
office:  MIT Dept. of EAPS 54-1424;  77 Massachusetts Ave.
             Cambridge, MA 02139-4307
emails:  eh3 at mit.edu                ed at eh3.com
URLs:    http://web.mit.edu/eh3/    http://eh3.com/
phone:   617-253-0098
fax:     617-253-4464