Providing quality of service in the Internet.

摘要

This dissertation presents a framework for providing quality of service (QoS) in the Internet. This framework consists of traffic directing and load balancing at the application layer, differentiated services (Diffserv) at the transport layer, traffic engineering and fast reroute at the network layer.;Traffic directing is to utilize the high-performance part of a network as much as possible and avoid using the low performance part. Load balancing is to distribute client requests to multiple servers so that service availability and responsiveness is increased. Approaches for traffic directing and load balancing are briefly described.;Diffserv is to divide traffic into different classes and treat them differently, especially when there is a shortage of network resources. Mechanisms needed at the edge and at the core of the network are described. These mechanisms can be used to achieve the desired per-hop behaviors (PHBs). By concatenating all these PHBs together, a certain level of QoS can be provided end to end. A conflict between Diffserv and TCP is also described, and a solution for resolving this conflict is proposed. This solves a significant problem in the Internet.;Traffic engineering is an iterative process of network planning and network optimization. Network planning is to improve the architecture (topology and link capacity) of a network in a systematic way so that the network is easy to operate, robust, and adaptive. Network optimization is to control the mapping and distribution of traffic over the existing network infrastructure to avoid and/or relieve congestion, and thus to optimize resource efficiency. The issues of designing a traffic engineering systems are discussed. A national traffic engineering system with multi-protocol label switching (MPLS) is then presented, and its performance evaluated. Based on the experience of implementing this system, a generic procedure is proposed for deploying large-scale traffic engineering systems. An approach for performing inter-domain traffic engineering in a quantitative way is then described. We also propose an offline constraint-based routing algorithm for computing the paths for MPLS label switched paths (LSPs), and describe how to provide Diffserv in an MPLS environment. Two emerging technologies that are closely related to traffic engineering, multi-protocol lambda switching (MPLmS) and fast reroute , are also discussed.;Constraint-based routing is one of the most important tools for traffic engineering. The issues related to constraint-based routing are discussed in detail. Heuristics for computing paths with propagation delay constraint, and for QoS routing on the top of virtual networks constructed in the traffic engineering process, are proposed. In order to reduce the computation complexity of constraint-based routing, an algorithm is proposed for reducing the routing table computation cost for OSPF. The correctness of the algorithm is illustrated. Similar idea can also be applied to IS-IS.;In summary, this dissertation not only discusses on all major issues related to QoS, but also presents a systematic approach for providing QoS in the Internet.