On the improvement from scheduling a two-station queueing network in heavy traffic

Abstract

For a two-station multiclass queueing network in heavy traffic, we assess the improvement from scheduling (job release and priority sequencing) that can occur relative to Poisson input and first-come first-served (FCFS) sequencing. In particular, simple upper bounds are derived on the optimal objective function value (found in Wein [7]) of a Brownian control problem that approximates (via Harrison's [2] model) a two-station queueing network scheduling problem in heavy traffic. When the system is perfectly balanced, the Brownian analysis predicts that optimal scheduling will reduce the long run expected average number of customers in the network by at least a factor of four relative to the Poisson input, FCFS sequencing policy that achieves the same throughput rate. When the system is not perfectly balanced, the corresponding factor is slightly smaller than two.