Concurrent gradient estimation for multiclass networks via ersatz estimators

Ersatz estimators are indirect measures of derivatives of performance functions. In this paper we apply them to a telecommunications queueing network. Through these estimators, the derivatives with respect to call arrival rate, cell arrival rate, and polling discipline at the server can all be recovered from a single simulation-based statistics. We present the theoretical basis for the transformations and verify their validity through computer simulations. Our methodology can be generalized and may yield the basis for efficient implementation of online flow, call admission, routing and polling control to achieve more intelligence in a distributed environment.