Search

This report documents an experiment investigating the potential of a parallel computing architecture to enhance the performance of a knowledge-based signal understanding system. The experiment consisted of implementing and evaluating an application encoded in a parallel programming extension of Lisp and executing on a simulated multiprocessor system. The chosen application for the experiment was a knowledge-based system for interpreting pre-processed, passively acquired radar emissions from aircraft. The application was implemented in an experimental concurrent, asynchronous object-oriented framework. This framework, in turn, relied on the services provided by the underlying hardware system. The hardware system for the experiment was a simulation of various sized grids of processors with inter-processor communication via message-passing. The experiment investigated the effects of various high-level control strategies on the quality of the problem solution, the speedup of the overall system performance as a function of the number of processors in the grid, and some of the issues in implementing and debugging a knowledge-based system on a message-passing multiprocessor system. This report describes the software and (simulated) hardware components of the experiment and present the qualitative and quantitative experimental results. (Author)

This paper describes a protocol to support point-to-point interprocessor communications with multicast. Dynamic, cut-through routing with local flow control is used to provide a high-throughput, low-latency communications path between processors. In addition, multicast transmissions are available, in which copies of a packet are sent to multiple destinations using common resources as much as possible. Special packet terminators and selective buffering are introduced to avoid deadlock during multicasts. A simulated implementation of the protocol is also described., Prepared in cooperation with Digital Equipment Corporation, Maynard, MA.

Simulation of systems at an architectural level can offer an effective way to study critical design choices if (1) the performance of the simulator is adequate to examine designs executing significant code bodies -- not just toy problems or small application fragments, (2) the details of the simulation include the critical details of the design, (3) the view of the design presented by the simulator instrumentation leads to useful insights on the problems with the design, and (4) there is enough flexibility in the simulation system so that the asking of unplanned questions is not suppressed by the weight of the mechanics involved in making changes either in the design or its measurement. A simulation system with these goals is described together with the approach to its implementation. Its application to the study of a particular class of multiprocessor hardware system architectures is illustrated.