NeTS-NR: Generalizing the Network Layer

Although the Internet has scaled far beyond the dreams of its designers, the service it provides represents but one point in a space of possible mechanisms and tradeoffs. As a result, systems that offer alternative network-type services have received a good deal of attention in recent years; examples include computed-routing services, distributed hash tables, indirection services, and publishsubscribe services. However, essentially all such proposals assume the existence of an underlying network service (i.e. the Internet), and are constructed in the form of an overlay. This has the disadvantage that routing functionality is duplicated; moreover, efficiency of routing in the overlay network does not, in general, correspond to efficiency in the underlying routing system. This project takes a fundamentally different approach to enhancing network-layer services, by considering alternative points in the network-layer design space. The vehicle for this investigation is a novel network service called Speccast. In the speccast service model, each packet carries a predicate (on nodes); the network is responsible for delivering the packet to all nodes that satisfy that predicate. This very general service model subsumes traditional unicast and multicast, and also supports enhanced services such as mobility and publish-subscribe, essentially for free. The main research challenges for this project are: 1. To understand the fundamental tradeoffs between flexibility and scalability in routing, addressing and forwarding. Building on a preliminary investigation, we will compare solutions and quantify the cost/benefit tradeoffs inherent in each. 2. Developing a variety of implementation strategies for speccast, with the objective of supporting networks as large as the Internet while minimizing administrative costs. Intellectual Merit: This project has the potential to change the (implicit) assumptions that have constrained the networking community’s thinking over the past two decades of working with the Internet. An explicit goal of the project is to advance understanding of fundamental tradeoffs among locality and hierarchy of addressing, scalability, and flexibility. Broader Impact: The speccast service offers the ability to support a wide range of new and emerging services using a single, generalized, yet simple, service abstraction, which could easily be incorporated into future routers, subsuming the existing IP routing services. The proposed work will also have a significant educational component. In addition to the usual graduate student research problems, we expect to integrate the new insights and understanding of fundamental routing issues produced by this project into our classes in the form of new courses and materials to be shared with other instructors and integrated into the broader networking curriculum.