Parallelizing ROMS for Distributed Memory Machines using SMS
Daniel S. Schaffer
NOAA/ERL/Forecast Systems Lab
Boulder, USA
The National Oceanographic and Atmospheric Administration Forecast
Systems Laboratory (FSL) has developed a directive-based tool
for parallelizing weather and ocean models. The user inserts
directives in the form of comments into existing FORTRAN code.
SMS translates the code and directives into a parallel version that
runs efficiently on both distributed and shared memory machines.
Here we describe the SMS tool and show how it has been used to
parallelize a Northeast Pacific scenario of the ROMS model. The
parallel code is currently being run in production on FSL's cluster
of Alpha workstations. Analysis of the parallel scalablity is given.
Plans to handle distributed memory parallelization of a nested ROMS
and a coupled COAMPS/ROMS model are discussed.