Modelling and performance analysis of inter-processor messaging in distributed systems

This paper addresses some performance modelling issues arising from the inter-processor messaging requirements of distributed real-time processing systems. We consider two basic classes of message transfer protocols, namely clocked schemes where message transfer is initiated as a periodic timed task, and event-driven schemes where the transfer mechanism is triggered by the messaging requests themselves. The aim is to maximize system efficiency by passing messages in batches. It is shown how the classes of protocols may be modelled by a two-stage queueing systems, which is analysed using the theory of imbedded Markov chains and semi-Markov processes. The results are used to show how the important performance measures are derived, and how the protocol parameters should be chosen to optimize the overall message system performance. The methods are illustrated by numerical examples.