[MITgcm-devel] Re: [MITgcm-cvs] MITgcm/verification/fizhi-cs-32x32x10/input CVS Commit

[Andrea Molod]

hi ed,

right. i understand all this. but removing the "useSETRLSTK=.TRUE."
flag from this experiment will at least let us see if the next
fizhi experiment (aqualev) works. and if it does not, i will remove
it from the aqualev too. just to let the testing on itrda proceed.

the 'advantage' of keeping it there so that these experiments can
run on faulks under g77 is not a huge one - the answers are wrong
for the 'full' fizhi experiment there anyway because think that
the sequential access reads of big endian files are still not right.
the model does not croak with an 'end of record' anymore but they are 
not reading things right, or at least that is my impression.

thanks,

andrea

On Fri, 16 Sep 2005, Ed Hill wrote:

> On Fri, 2005-09-16 at 16:04 -0400, Andrea Molod wrote:
> > Update of /u/gcmpack/MITgcm/verification/fizhi-cs-32x32x10/input
> > In directory forge:/tmp/cvs-serv8436
> > 
> > Modified Files:
> > 	eedata 
> > Log Message:
> > comment out SET_RLSTK which may be creating problems on itrda
> 
> Hi Andrea,
> 
> Technically, "useSETRLSTK=.TRUE." isn't a root cause.  All that
> useSETRLSTK does is un-limit the stack size.  Its identical to:
> 
>   [ba]sh:  ulimit -s unlimited
>   [t]csh:  limit stacksize unlimited
> 
> It was the stack size limit that killed fizhi w/ g77 on those machines
> (and, earlier, with the Intel compilers on Columbia).  The useSETRLSTK
> flag now allows the run to proceed a bit further along.  In this case,
> it encounters a different problem that results in the test taking longer
> to execute than the wall time allowed by our queue.  So, the entire set
> of tests times out and we get no results.
>