We present a parallel Poisson solver on distributed computing environments. In the solver, the parallel implementation of the Fast Multipole Method (FMM) is designed to minimize amount of data communication and the number of data transfers and synchronizations. The experimental results show linear speedup, good load balancing, and reasonable performance under failure and demonstrate the viability of loosely coupled heterogeneous workstations for large scale scientific computations.
Bibliographical noteFunding Information:
This work was partially supported by Ewha Womans University Research Grant, 1996 and by Korea Science and Engineering Foundation, KOSEF:970701-01013. We thank S. Talht for his PLinda coding help in early stage of our work, and F. Ethridge, L. Greengurd, and D. Shasha for reading our l~per and giving many valuable comments. The experiments have been done at the Courant Institute of Mathem~ical Sciences (CIMS) of New York University with the support of the Courant Mathematics and Computing I~boratory (CMCL) and the Department of Computer Science.
- Adaptive quad-tree
- Domain decomposition
- Fast direct Poisson solver
- High order of accuracy
- Volume integral method