The objective of this project is the study and development of hardware and software for performance analysis of advanced multiprocessor systems. The work is centered around the Cedar multiprocessor project, partly funded under another NSF grant. The intent is to develop measurements and analytic tools that can be applied to the performance evaluation of the Cedar machine as well as Alliant FX/8. These include static analysis tools to address the algorithm/architecture mapping issues related to the efficient use of multiprocessors. It also includes the development of instrumentation for dynamic analysis; namely, hardware and software for collecting run-time execution data from the Cedar system. Another related goal is to investigate several techniques for virtual execution from analysis of multitasked programs on Cedar. The evaluation of parallel machine performance is in its infancy and it is likely that the field may benefit from the work proposed. This is one of the strongest research groups in parallel processing in the country. Their track record is impressive. The funding of this work at reduced level is strongly recommended.
