next up previous contents
Next: 6.2.1 Workload Distribution Up: 6. Parallel and Distributed Previous: 6.1.3 Summary

6.2 Distributed Computing

As long as the parallel computation is not carried out on a single machine with multiple processors, a simulation has to spread out to remote machines using a network connection as depicted in Figure 6.3. It will therefore use machines with different properties in terms of speed, memory, and system load. This implies that the computation time which is necessary to finish a given simulation problem will differ from one machine to another. A further implication of distributed computation is an increased susceptibility to single point failures, which will degrade the robustness of the whole simulation system unless measures are taken to account for them.

% latex2html id marker 6442\centering
Distributed computing on a cluster
of heterogeneous workstations.}

Rudi Strasser