Study On The Overlay-Based Internet Infrastructure

As the Internet technology being accepted widely, computer networks become increasingly influential on human lives. A variety of novations in the Internet applications are emerging rapidly. In contrast, today's Internet infrastructure almost remains the same as what it was in essense. As the Internet scales up and the novel applications come out continually, the inadequacy in the old infrastructure is showing up. For example, the BGP routing algorithm has a lot of problems such as slow recovery and path inflation, which could impact the end-to-end transmission seriously. The lack of support for the multicast and the QoS guarantees makes it difficult to implement applications, which have common requirements in these fields. On the other hand, people are accustomed to seperate the network infrastructure from the end hosts. They believe that the infrastructure should provide the access and transmission service, while the end hosts are only responsible for the execution of applications. In this model, we have to substitute all network devices, like routers, to upgrade the infrastructure, which is very costly. As the prospect of the applications is not clear, network operators are unwilling to take the risks. This situation also retards the development of the network infrastructure.Overlay networks make people consider the Internet infrastructure from a new perspective. An overlay network consists of a small portion of nodes from an underlying network for a specific purpose. In the past few years, the application of overlay networks, such as resilient overlay network (RON) and overlay network multicasting, has become a hot topic in the Internet society. Although most of these overlays are built over end hosts at the application layer, they try to provide basic services like routing and multicasting, which was believed to be the duty of the network infrastructure. Therefore, we believe that all kinds of overlays can be important components of the Internet infrastructure, while the overlay technology can be a necessary mechanism for us to change and upgrade the Internet infrastructure. Our view is based on these two points: first, a lot of overlay networks have the same purpose as the nework infrastructure based on routers so that the former can be an effective complement of the latter; second, an important idea in the overlay technology is to provide the basic service by a small number of nodes, which saves up the huge cost to change the underlying network thoroughly and offers good flexibility. According to our knowledge, overlay networks can be a good supply to the current Internet infrastructure in improving the routing quality and in providing multicast service. However, there remains a lot of problem to solve if we want to merge the overlays into the network infrastructure. This paper will focus on these questions.The deployment problem of routing overlay network. Routing overlay network is designed to improve the routing service provided by the IP layer. According to the analysis of the BGP routing protocol, we find that the BGP routing strategy uaually leads to path inflation, which could cause more path failures and bandwidth consumption. What's more, the path inflation mainly results from commecial aggreements between network operators, which is not purely a technical problem. The routing service through overlay network can help us to circumvent this problem. To form a routing overlay network, we have to find out a suitable deployment scheme, which should be not only economic but also practical. Chapter 2 is devoted to this problem.The construction of the topology of dynamic overlay networks. We use the term dynamic overlay network to denote those overlays, in which each nodes is not only a data consumer but can also relay data to others. Such overlays are usually employed to provide multicast or content distribution service. A key problem for such overlays is how to organize those dynamic members into an overlay tree and to keep it fit the underlying network topologically. It is tightly related to the transmission efficiency and the stability of the overlays. In chapter 3, we propose a mechanism, based on two novel primitives, to make the overlay topology consistent with the underlying network. In addition, these two primitives can be supported by ordinary multicast routers.QoS guarantees in multicasting. The control of QoS is an important topic in the Internet society. Although there have been a large amount of reaches in this field and several practical products have come into beings, most of them are limited to unicast transmission. Since the overlay multicast session can be broken into the combination of unicast connections, we can depend on existing QoS control mechanisms to offer QoS guarantees in overlay multicasting. In chapter 4, we propose a scheme to provide QoS-aware multicast through the famous Diffserv framework.