Optimizing communication networks: Incentives and multipath transfers

The past ten years have seen dramatic changes in networking. For example, wireless and the Internet have overtaken traditional telecommunication networks. Yet some of the economic incentives and standardspsila policies that have enabled such growth have also inhibited other forms of evolutionary development. For instance, the fundamental architecture of the Internet has changed little. This has led to flurry of interest in ldquoClean Slaterdquo approaches to next-generation networks. Instead, we ask how far it is possible go without re-architecting networks, by exploiting some current research ideas.We discuss control strategies for communication networks such as the Internet, or multihop networks where the aim is to optimize network performance in some sense. We address some key issues concerned with economics, bandwidth guarantees and security, using content distribution as a motivating example. We describe how welfare maximization can be used as a paradigm for network resource allocation, and also be used to derive practical rate-control algorithms. We apply this to the case of multipath transfers, where multiple paths are used to transfer data, and where we can use decentralised protocols to effectively load-balance across the network. We discuss how this, coupled with a dynamic path selection scheme can lead to an efficient ldquosocial welfarerdquo optimum, and how some current peer-to-peer systems embody elements of this approach. The benefit to the users is better performance, while offering simpler and more robust engineering for ISPs. We comment on incentives and pricing issues for both users and providers.