This award will support a joint effort research in the evaluation of parallel processing systems, to be carried out by researchers based at the University of Massachusetts and the French National Institute for Information Science and Automation (INRIA). The U.S. group will be led by Dr. Donald Towsley; on the French side the project will be led by Dr. Francois Bacelli, INRIA, Rocquencourt. The research project will focus on the following two problems: 1) The development of a mathematical methodology for evaluating the performance of complex parallel systems. There currently exist powerful tools for evaluating the performance of sequential programs that execute on single or multiprocessors. The main difference between a sequential and a parallel program is the need for synchronization of the different activities present in a parallel program execution. Consequently, the first goal of the proposed research is to develop a methodology for evaluating parallel software applications that exhibit this phenomenon. 2) The design and analysis of schedulers for multiprogrammed multiprocessors that execute parallel programs. Many existing multiprocessors have implemented scheduling policies originally developed for single processors. Initial studies indicate that this can result in poor scheduling polices, i.e., policies that yield poor performance. The second goal of the proposed research is to develop a methodology for designing scheduling policies that yield good performance, such as low response times. This research joins two groups of outstanding U.S. and French computer scientists. INRIA is widely recognized for its concentration of expertise in computer research, automation, mathematics and scientific computing. It provides an excellent combination of fundamental research with industrial collaboration in these areas.
