Adaptive source routing of real-time traffic in integrated services networks

Increasing bandwidth is enabling the deployment of real-time applications in packet-switched computer networks. The stringent performance requirements of these applications necessitate a reexamination of the fundamental architectural components of today's networks, which were designed to support traditional elastic applications. Network routing is one such component. In this thesis, we undertake a thorough examination of how routing should operate in future integrated services networks, which will support real-time applications along side elastic ones. We propose a new routing architecture to support real-time traffic and study the performance of this architecture using simulation. Our approach to routing, which we call Adaptive Source Routing, includes the following key features. Multiple paths between each source and destination are computed using a static metric. These include both minimum cost paths as well as longer alternate paths. The source of traffic selects routes using dynamic information that is distributed to network nodes. These features enable the routing mechanism to locate quickly routes that meet the performance requirements of real-time applications. Source routing facilitates the use of multiple paths in large networks by preventing routing loops without requiring all network nodes to maintain consistent information. We study the performance of Adaptive Source Routing using simulation. We find that when compared to other algorithms, it increases the amount of real-time traffic a network can carry and decreases the delay a real-time session encounters when it is initiated. We show how the magnitude of these benefits depends on network topology and traffic patterns. We also consider issues specific to routing in large networks or internets, focusing on the overhead of dynamic information distribution. The frequent global distribution of dynamic information in a large network is not feasible. Therefore, we explore methods for the limited distribution of dynamic information and for adaptive routing without the distribution of any dynamic information. Using simulation, we find that over a range of network topologies and traffic conditions, these mechanisms significantly reduce the overhead of update distribution relative to global flooding while still providing some of the benefits that result from adaptive routing.