R. F. DeMara, "Characterizing Marker-Propagation Mechanisms," in Proceedings of the 
First Workshop on Abstract Machine Models for Highly Parallel Computers, pp. 77 - 82, 
Leeds, United Kingdom, March 25 - 27, 1991.

Abstract:
The need for fast, scalable, and efficient computer architectures for Artificial 
Intelligence(AI) applications is well recognized. However, relatively little is known about 
the performance of parallel AI architectures. In this paper, we present techniques to 
evaluate and improve marker-propagation architectures which utilize the massive 
parallelism inherent in many AI applications. Based on an analysis of marker-passing 
programs and knowledge based, we developed a set of performance indices by defining 
concepts such as marker Power and Dispersion. These indices and a classification of 
workloads were used to construct a suite of performance benchmarks. We then developed 
a 160-processor marker-passing super computer called SNAP-1 which was tailored to 
provide visibility into the performance-critical features of marker-propagation 
architectures. Finally, we devised a hybrid hardware/microcode tracing methodology to 
collect and interpret the results in terms of the metric we have defined.