PERFORMANCE ANALYSIS OF PARALLEL OCEAN
MODELS USING HPF AND MPI
S. Piacsek, Code 7320, Naval Research Laboratory
Stennis Space Center, MS, 39529, USA
M.Young, Code 5593, Naval Research Laboratory
Washington, DC, 20375
W.Oberpriller, LCSE, Univ. of Minnesota,
Minneapolis, MN, USA
The scalability and on-processor performance of two ocean
models and one ice model were investigated using HPF and
MPI for parallel tools. The ocean models were the POM
(Princeton Ocean Model, and DieCAST , using sigma- and
z-coordinates in the vertical, respectively. The ice model was
a Hibler-type with the VP (viscous-plastic) ice rheology.
For resolutions of order 1000x500x30, the HPF version
of POM matched the MPI version on 128 processors. It
did not scale as well for larger number of processors. At
64 processors HPF was twice as fast as MPI on the T3E.