Next: : REFERENCES
Up: Clapp and Sava: Cluster
Previous: FUTURE
In this paper we summarized our computer building experience.
We found the most cost effective processor was a dual P3
with SDRAM. We found that MPI provides the best scalability, but the
standard implementation's lack of fault tolerance is problematic.
In the future we see using/developing a fault tolerant MPI which
allows for dynamic change in the number of nodes used by a given process.
Stanford Exploration Project
6/8/2002