Comparison of control schemes for a class of distributed systems

We consider a class of systems consisting of a main processor and subtending peripheral processors and study classes of control policies for limiting traffic to and the response time of the main processor. The classes correspond to window schemes, which control the load based on the number of outstanding customers in suitability defined stages of processing, and a variety of buffering disciplines designed to improve overload performance. Queueing network and Markovian models are used to analyze performance. We study the effect of location of the control point (centralized and various distributed schemes), as well as the effect of the queueing disciplines used, by comparing their throughput response time tradeoff characteristics.