MPI: a message passing interface
Proceedings of the 1993 ACM/IEEE conference on Supercomputing
Parallel Computing - Special issue: climate and weather modeling
A Fast and High Quality Multilevel Scheme for Partitioning Irregular Graphs
SIAM Journal on Scientific Computing
A conservative finite-volume second-order-accurate projection method on hybrid unstructured grids
Journal of Computational Physics
Predictive performance and scalability modeling of a large-scale application
Proceedings of the 2001 ACM/IEEE conference on Supercomputing
The Parallelization of the Princeton Ocean Model
Euro-Par '99 Proceedings of the 5th International Euro-Par Conference on Parallel Processing
A Performance Model of the Parallel Ocean Program
International Journal of High Performance Computing Applications
A simple parallelization technique with MPI for ocean circulation models
Journal of Parallel and Distributed Computing
| Hi-index | 0.01 |
The Finite Volume Coastal Ocean Model (FVCOM) is a publicly available software package for simulation of ocean processes in coastal areas. The unstructured grid approach used in the model is highly advantageous for resolving dynamics in regions with complex shorelines such as estuaries, embayments, and archipelagos. A growing user community and a demand for large-scale, high resolution simulations has driven the need for the implementation of a portable and efficient parallelization of the FVCOM core code. The triangular grid approach used in FVCOM precludes the utilization of schemes used previously in the parallelization of popular structured grid ocean models. This paper describes recent work on a SPMD parallelization of FVCOM. The METIS partitioning libraries are employed to decompose the domain. Parallel operations are programmed with the Message Passing Interface (MPI) standard interface. Updates for flow quantities near the interprocessor domain boundaries are performed using a mixture of halo and flux summation approaches to minimize communication overhead. Evaluation of the implementation efficiency is made on machines comprising several parallel architectures and interconnect types. The implementation is found to scale well on medium-sized (~ 256 processor) clusters. An execution time model is developed to expose bottlenecks and extrapolate the performance of FVCOM to increasingly available large MPP machines. Application to a model of water circulation in the Gulf of Maine shows that the parallelized code greatly increases the capabilities of the original core scheme by extending practical model simulation timescales and spatial resolution.