[MITgcm-support] Poor cpu usage percentage

Rogan, Nicholas nrogan at geomar.de
Mon Jul 20 10:47:59 EDT 2015


hi Dimitris,

Thanks for your patience! 

It appears that it is a problem with the I/O routines as there are significant differences in wall clocl times between the two compilations. 

The main routines that are time expensive are 
LOAD_FIELDS_DRIVER
>> EXF_GETFORCING 1500s vs 300s

MONITOR 1700s vs 300s 

and 

DO_THE_MODEL_IO 1900s vs 1200s

There aren't really any other major differences, in fact on the whole the users and system times for the iMac are less than those for the macbook.

Looking at the Makefiles there are the following differences:
iMac
- Does not have a specified F90C compiler (The macbook pro has gfortran specified as F90C compiler)
 NB The other compilers are identical between the two compilations, i.e. FC=gfortran, CC=gcc, LINK=gfortran

FFLAGS (relative to macbook)
- Does not have: -Wunused
- Does have : -f-no-second-underscore -fbacktrace
- BOTH HAVE: -Wunitialized -fconvert=big-endian

FOPTIM
- Does not have: -ftree-vectorize

NOOPTFLAGS
- Does not have: -O2 -funroll-loops
- Does have : -O0

LIBS
- Does not have: -lcurl

There are also a lot of included .h files that occur in the MacBook's Makefile that are absent from the iMac.

Any quick suggestions? I'll keep work myself on trying to figure out the implications of the differences.

Thanks!

Nick



On 20/07/15 03:02 PM, Dimitris Menemenlis  <dmenemenlis at gmail.com> wrote:
> 
> 
> 
> 
> 
> you mainly need to look at your fortran and c compilers and compiler options.
> 
> 
> 
> 
> > 
> > On Jul 20, 2015, at 9:29 AM, Rogan, Nicholas <nrogan at geomar.de> wrote:
> > 
> > 
> > I'm comparing the two Makefiles now, so far only different Netcdf/compiler locations.
> > 
> > 
> 
> 
> 
> 
> 
> 
> 
-- 
Kind regards,

Nick

------------------------------------------------------
Dr Nicholas Rogan
Postdoctoral Researcher
http://www.geomar.de/en/mitarbeiter/fb2/ch/nrogan/

GEOMAR Helmholtz Centre for Ocean Research, Kiel
Wischhofstraße 1-3, Build. 12
D-24148 Kiel
Germany
------------------------------------------------------
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mitgcm.org/pipermail/mitgcm-support/attachments/20150720/ce0fb6a4/attachment.htm>


More information about the MITgcm-support mailing list