Parallelizing a new class of large applications over high-speed networks

parallel architecture. In both cases, the efforts require Several large applications have been parallelized that the persons responsible for the porting are on Nectar, a network-based multicomputer recently intimately familiar with the applications. Since large developed by Carnegie Mellon. These applications applications tend to involve substantial applicationwere previously either too large or too complex to be specific knowledge, this implies that often only easily implemented on distributed memory parallel application scientists themselves can do the porting. systems. Parallelizing these applications was made possible by the cooperative use of many existing This definitely is not the best way of using their time, general-purpose computers over high-speed and as a result, many large applications have never been networks, and by an implementation methodology implemented on parallel machines. based on a clean separation between applicationspecific and system-specific code. We illustrate What we need are parallel architectures and these points using our experience with parallelizing programming tools to provide direct and general three real-world applications. The success in these support for parallelizing large applications. These tools applications clearly points out a new direction in parallel processing. should be at a higher level than send and receive primitives, or synchronization or shared data primitives. The objective is to be able to port large applications