9504291 Schauser The object of this proposal is to study parallel algorithms that run efficiently on modern multiprocessors and, in particular, on networks of workstations (NOWs). This research will use a simple but realistic model of communication, the LogP model. The goal is to use the model and its parameters to predict and improve the performance of algorithms on various architectures. In particular, it will provide a methodology for porting efficient code from one machine(e.g., a low-end NOW) to another (e.g., a high-end NOW/MPP), such that both versions are efficient on their respective architectures. This work will be mostly of an experimental nature, and will be carried out on the Meiko CS-2. Some benefits to this research are : accurately predicting the performance of parallel code on other architectures, efficiently optimizing the ported algorithms, and making the use of low-end NOWs as test platforms much more viable.
