Understanding how to provide service guarantees in IP-based networks

The growth of IP networks has obviously been accompanied by the tremendous growth not only in their sizes and the volume of traffic they carry, but also in the range of requirements they are expected to meet. On one hand, this has brought renewed interest in Quality-of-Service (QoS) solutions, i.e., solutions that enable service differentiations. On the other hand, the ever increasing speed and traffic volume of IP networks, particularly in backbone networks, have made the traditional alternative to QoS, namely overprovisioning, more favorable from a complexity perspective. Understanding how to provide service guarantees in IP networks has been the topic of a long-lasting debate within the networking research community. The goal of this thesis is to provide a better understanding of the roles and potential of both over-provisioning and QoS in such a setting. First, we investigate the extent to which the ever increasing size of IP networks can play a positive role in their ability to absorb traffic variations. We develop a general model that accounts for network topology, base offered traffic, and traffic variations, and allows us to explore how their combination behaves as the network grows. We identify critical thresholds in the relation between network and traffic growth, which delineate regions where a small amount of over-provisioning can provide increasing protection against traffic variations. The results offer insight into how to grow IP networks in order to enhance their robustness. Second, we study the feasibility of providing meaningful service guarantees without introducing too much added complexity. We propose two QoS schemes that can provide iv stronger and more flexible service guarantees. Our first scheme provides differentiated loss guarantees by adding a simple random drop logic on top of the basic FIFO queue. Our second scheme provides more flexible support for real-time applications without hard-limiting their traffic to traffic contracts, and we achieve this goal through a combination of scheduling and buffer management mechanisms. The performances of our schemes are shown to achieve our design goals through simulations or actual implementations