Nesting ocean model for parallel vector processors
Yasumasa Miyazawa
Institute for Global Change Research/ FRSGC, Japan
Akihiro Musa
NEC Corporation, Inc., Japan
Koji Ogochi
Information Technologies of Japan, Inc., Japan
We developed a nesting ocean model based on the Princeton Ocean Model
for parallel vector processors to perform predictability experiments of
the Kuroshio path variation south of Japan. The meso-scale eddy activities
are indispensable factors for the Kuroshio path variation (Miyazawa et al,
2001). The horizontal/ vertical resolution significantly affects
representation of meso-scale eddy activities. Moreover the coastal
application of the model results needs the nesting method; embedding
higher resolution model in the present version of the model. In order
to satisfy the above demands, a one-way nesting ocean model in which
two models on different nodes are connected by MPI is developed. The
higher resolution model itself is parallelized on different processor
nodes by MPI using the domain decomposition method for reasonable load
distribution. The part assigned to each node is properly vectorized for
effective computation on the architecture of vector-parallel processors.