Timeout Control In Distributed Systems Using Perturbation Analysis

Timeout control is a simple mechanism used when direct feedback is either impossible, unreliable, or too costly, as is often the case in distributed systems. Its effectiveness is determined by a timeout threshold parameter and our goal is to quantify the effect of this parameter on the behavior of such systems. In this paper, we consider a basic communication system with timeout control, model it as a stochastic flow system, and use Infinitesimal Perturbation Analysis to determine the sensitivity of a "goodput" performance metric with respect to the timeout threshold parameter. In conjunction with a gradient-based scheme, we show that we can determine an optimal value of this parameter. Some numerical examples are included.