A Experimental Study of Alternative Schemes for Asynchronous Message Passing in a Real-Time Multicomputer Control System.

This dissertation is concerned with the problem of communication between multiple computers for applications involving the real-time control of a complex physical system. The class of physical objects selected for experimentation is that of multi-legged robots or "walking machines". The computers used are commercially available single-board systems. Six of these boards are connected to a common bus to form the multicomputer system which is the hardware object of this dissertation. The walking machine to be controlled by the multicomputer is simulated by one of the computer boards. The principal issue explored in this research is that of determining the most effective communication protocol for the available multicomputer relative to the selected application. Three alternative protocols are proposed: namely; (1) asynchronous multiple bus-master, (2) polling single bus-master, and (3) interrupt-driven single bus-master. For each of these protocols, commands as defined as messages which must be consummed exactly once while data defines messages which may be consumed any number of times, but must always be in freshest form. For both commands and data, but communication between computer boards is assumed to be fully asynchronous. In order to carry out the comparison of the three proposed protocols, the Run Time Support software provided by the manufacturer was rewritten to operate in a multicomputer environment. In addition, physical properties of the bus were determined by experimental means to provide a detailed understanding of the time delays and processing steps associated with cross-bus communication. The major result of this investigation is an experimental proof that the multicomputer approach is effective for the selected application and that all three protocols tested are satisfactory, although the multiple bus-master approach gives the best performance. This result is due to a low level of bus traffic resulting from careful partitioning of the control problem into processes requiring relatively little communication. In a higher bus traffic situation, it is possible that some other protocol would be preferable.