QoS routing: challenges and solution approaches

Summary form only given. Broadband integrated services networks are expected to support multiple and diverse applications with various quality of service (QoS) requirements. Accordingly, a major issue in the design of broadband architectures is how to provide the resources in order to meet the requirements of each connection, and, moreover, how to meet this goal in a networkwide efficient manner. QoS routing is, undoubtedly, one of the major building blocks in such architectures. However, QoS routing poses several major algorithmic challenges. One complication is the inherent intractability of many fundamental QoS routing problems. A second challenge is the need to provide scalable solutions, which can cope with the growing sizes of networks. An additional complication lies in the typical uncertainty regarding the precise state of the network. All these become ever more challenging when addressing multicast connections. Moreover, a successful QoS routing scheme should smoothly integrate with other QoS-enabling mechanisms, most notably schedulers. Finally, the practical deployment of QoS routing requires to successfully integrate it within standard routing protocols. These major challenges call for novel algorithmic approaches and solution schemes, which are the subject of this talk. We shall overview some typical problems and solution approaches, focusing on three representative models. The first is a "basic" model, where each network element can offer a certain degree of QoS at a certain "cost". In the second, "extended" model, each network element may offer various degrees of QoS, at different "costs". Finally, in the "coupled" model, the tasks of routing and scheduling are tackled together, hence providing a more precise assessment of the actual consumption of resources. We shall consider the support of bottleneck and additive QoS requirements, both for unicast as well as multicast connections, also in the presence of network failures. We shall focus on solution schemes that provide proven performance guarantees within efficient (polynomial) time complexity for general network topologies. We shall overview several approaches for improving scalability, with a particular focus on precomputation schemes. We shall indicate how this algorithmic framework can address problems of special practical interest, such as the need to cope with state uncertainty