On-line minimization of call setup time via load balancing: a stochastic approximation approach

With the addition of new network services, it is anticipated that the processing involved in setting up a call in a circuit-switched network or a session in a packet-switched network will vary greatly for different types of services. In this paper, we address the problem of reducing the call setup time in a circuit-switched network, or equivalently the session setup time in a packet-switched network, through the balancing of load across call processors. With a view to designing algorithms to execute on-line in a system, we formulate a stochastic optimization problem and study the use of stochastic approximation techniques. Given the distributed nature of the problem, we extend previous results obtained for a single node to the case where several nodes operate simultaneously and in an asynchronous manner. Our results include a theoretical study of convergence as well as several simulation results that compare two stochastic approximation techniques