[MITgcm-devel] SERIOUS performance problems with latest code

[Patrick Heimbach]

Alistair,

yes, I know.
At GFDL I just ran an older version (c58d)
using latest compiler successfully (i.e. expected perf.).
Baylor ran latest code at GFDL with bost
new and old compiler setting with low perf.

At NASA Ames I ran c58d with both their
old and new compilers successfully.
I ran latest model with both old and new opts'
with bad perf.

So bottom line seems,
not a platform/disk problem,
not a compiler problem.

-p.



On Tue, 2006-10-31 at 10:48, Alistair Adcroft wrote:
> Patrick,
> 
> On what systems are you finding performance problems? At GFDL, there's 
> been a major overhaul of OS's and there's been a backward compatibility 
> problem with object files, executables and environment variables.
> 
> In your scripts for compiling and running you need to ensure you have 
> these modules loaded:
> 
>  module purge
>  module load icc.9.1.042
>  module load ifort.9.1.033
>  module load scsl-1.5.1.0
>  module load mpt-1.14
>  module load idb.9.0.12
>  module load totalview_730-0
> 
> I believe the build_opts file for GFDL has a libarary reference that 
> needs to be fixed also. With the above, particularly the mpt module, you 
> should see a performance improvement over previous runs on the same machine.
> 
> A.
> 
> Patrick Heimbach wrote:
> > It seems that the latest code has severe
> > performance problems.
> > My ECCO forward run takes 4 times the known
> > performance at NASA/Ames, both with old and
> > new optfiles/compilers.
> >
> > Baylor experiences similar problems,
> > getting only 20% CPU time per wallclock time
> > (expected value at GFDL is around 95%, I think).
> >
> > Anyone witnesses similar problems,
> > or has ideas of what changes in code could cause it?
> >
> > -Patrick
> >
> >   
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel
-- 
Dr Patrick Heimbach | heimbach at mit.edu | http://www.mit.edu/~heimbach
MIT | EAPS, 54-1518 | 77 Massachusetts Ave | Cambridge, MA 02139, USA
FON: +1-617-253-5259 | FAX: +1-617-253-4464 | SKYPE: patrick.heimbach