Subject: Problems solved with compiling/executing definsurf program on SGI Workstation
From: Aidan Heerdegen (aidan@modjadji.anu.edu.au)
Date: Thu Jan 21 1999 - 15:01:23 MST
Jim Rosinski <rosinski@bearmtn-e0.cgd.ucar.edu> wrote:
> On Tue, 19 Jan 1999, Aidan Heerdegen wrote:
>
> > Segmentation fault (core dumped)
> >
> > Definesurf seems to be writing the SGH and PHIS fields out with a 90 degree
> > shift in longitude i.e. the data for South America appears just east of Africa!
>
> Trying to make sense out of a netcdf file written by a job which was not
> run to completion (e.g. seg faulted) is likely to be a fruitless exercise.
> Since netcdf writes random access files with liberal use of buffering, there
> are no guarantees that calls to routines like "NF_PUT_VAR" actually result in
> data being disk-resident immediately after the call. Or even after the
> program terminates in this case, since the seg fault likely occurred *before*
> the NF_CLOSE call which would have flushed the buffers.
Exactly as you said. Once I got it running without the segmentation fault
the fields were fine.
> I don't know what resolution you are interpolating to, but one thing to try
> might be to interpolate to a very low resolution (e.g. 16x8) and see if that
> works. If it works, I would suspect a stacksize problem as the cause of the
> seg fault for the higher resolution case.
Well I tried that, and it still gave an error. This was due to the people
who run the SGI putting a *very* low limit on stacksize for interactive
jobs. I didn't think that such a small job would require queueing, so I
never tried it. Once I did, it was fine.
I thought I would share this with the group, in case others were using
shared machines with extensive NQS type queues. Perhaps their sysadmins
have done similar things with their systems and all they would need to
do is submit the definesurf job to the queue.
Thanks very much for the help.
As an aside - I tried to compile the tools on a largish Sun SPARC Ultra
and it didn't like the real*32 variables, and so didn't compile at
all. A similar result with my small Sun SPARC 5 workstation. I also
tried to compile on the Fugitsu VPP300 here at the ANU and it didn't
like it. The SGI Power Challenge compiled fine however. This is not a
criticism, just an observation about what sort of 'vanilla' systems
seemed to work out of the box, so to speak.
Cheerio
Aidan
-- Aidan Heerdegen Department of Geography aidan@dart.anu.edu.au Australian National University +61-2-6279 8150 ACT 0200, AUSTRALIA
This archive was generated by hypermail 2b27 : Thu Jun 01 2000 - 09:26:35 MDT