Abstract:
Stochastic queueing networks have become increasingly important in recent years as models of communication networks. There has been much interest in modeling and improving the control mechanisms for traffic flows through communication networks, which include both computer and transportation networks. The aim of this thesis is to study these two types of networks and hence to derive the system optimal controls. A new service control mechanism is proposed and applied to large parallel queueing networks which can be used as models for the Internet. A general framework and asymptotic optimal controls have been established for a particular type of large parallel queueing network. This thesis also analyzes in detail a queueing network with private transportation and public transportation in parallel which are modeled as an M/M/1 queue and a batch-service queue respectively. It is shown that the state-dependent system optimal policies have two monotone properties in the occupancy of these queues, and under the state-independent probabilistic system optimal policies there is no occurrence of Downs-Thomson paradox.