Reference: Noble, A. C. & Rogers, E. C. AIRTRAC Path Association: Development of a Knowledge-Based System for a Multiprocessor. June, 1988.
Abstract: Very Little is known about programming knowledge-based systems on multiprocessors. To understand the effectiveness of parallel implementations of such systems, programming problems and performance issues need to be studied at all levels of the computational hierarchy, from hardware to application. AIRTRAC is a simulated radar signal interpretation system for tracking and classifying aircraft, and runs on CARE, a simulated distributed-memory multiprocessor architecture. CARE consists of 1 to 1000 processor-memory pairs communicating via a network that provides reliable message delivery but without message ordering. AIRTRAC is implemented in LAMINA, an object-oriented applications programming interface to CARE, and ELMA, a high-level interface built on top of LAMINA. These are Zetalisp extensions which provide mechanisms and language interface syntax for expressing and managing concurrency. This report documents the development of the Path Association module of AIRTRAC, from design through implementation and testing, and describes the features of the supporting ELMA interface, which was developed in parallel with the application. We define the criterion of sustainable data rate for quantitative performance evaluation and describe experiments to determine performance under different conditions and their results. We discuss the techniques and constructs we used and the lessons we learned in the course of developing Path Association. We believe these lessons and results will be useful to others working in the field of parallel symbolic computation.
Notes: 101 pages.