A conceptual model for interprocess message-passing

The development of a general model for interprocess communication by message-passing is described. This development uses as its basis a compact set of communication semantic primitives, and concludes with the message-passing system represented by state machines defining a four-tuple state (the number of blocked processes sending messages, the number of blocked processes requesting messages, the number of messages pending delivery, and the number of requests for messages to be satisfied). The performance of a distributed system that relies on message-passing as the means of interprocess communication and synchronization can be evaluated using the models presented. Throughput and average queue lengths or growth rates are formulated in terms of input parameters that include the (constant) arrival rates of various communication commands, message processing architecture, and processing rates. Derivations based on approximations are used wherever a more rigorous approach leads to complex analyses. It is believed that the accuracy lost by approximation is a small price to pay for the simplicity gained. The model is validated by comparison of its results with those of a message-passing system simulator. Application of the model to the research distributed computing system AMOEBA, is illustrated, and recommended future research effort in the area is described.