Distributing a n-Body Problem Algorithm at Large-Scale over a Multi-Sites Grid Using JavaSpace
Abstract
This article presents the distribution of a n-body algorithm (a long-range data interaction application) using the JavaSpace technology (a tuple-based virtual shared memory) and reports the performance results observed when testing it at a large scale (hundredth of processors) on a multi-sites grid. It reveals that good speed-up may be acheived and provides a rule-of-thumb to choose the sensitive optimal number of processors to use. Scalability measurements show as well that the theoretical execution time of O(N2) can linearized to O(N) provided enough resources are available.