Looking Over the Fence at Networks: A Neighbor's View of Networking Research (2001)

The Internet has been highly successful in meeting the original vision of providing ubiquitous computer-to-computer interaction in the face of heterogeneous underlying technologies. No longer a research plaything, the Internet is widely used for production systems and has a very large installed base. Commercial interests play a major role in shaping its ongoing development. Success, however, has been a double-edged sword, for with it has come the danger of ossification, or inability to change, in multiple dimensions:

• Intellectual ossification—The pressure for compatibility with the current Internet risks stifling innovative intellectual thinking. For example, the frequently imposed requirement that new protocols not compete unfairly with TCP-based traffic constrains the development of alternatives for cooperative resource sharing. Would a paper on the NETBLT protocol that proposed an alternative approach to control called “rate-based” (in place of “window-based”) be accepted for publication today?

• Infrastructure ossification—The ability of researchers to affect what is deployed in the core infrastructure (which is operated mainly by businesses) is extremely limited. For example, pervasive network-layer multicast remains unrealized, despite considerable research and efforts to transfer that research to products.¹

• System ossification—Limitations in the current architecture have led to shoe-horn solutions that increase the fragility of the system. For example, network address translation violates architectural assumptions about the semantics of addresses. The problem is exacerbated because a research result is often judged by how hard it will be to deploy in the Internet, and the Internet service providers sometimes favor more easily deployed approaches that may not be desirable solutions for the long run.

At the same time, the demands of users and the realities of commercial interests present a new set of challenges that may very well require a fresh approach. The Internet vision of the last 20 years has been to have all computers communicate. The ability to hide the details of the heterogeneous underlying technologies is acknowledged to be a great strength of the design, but it also creates problems because the performance variability associated with underlying network capacity, time-varying loads, and the like means that applications work in some circumstances but not others. More generally, outsiders advocated a more user-centric view of networking research—a perspective that resonated with a number of the networking insiders as well. Drawing on their own experiences, insiders commented that users are likely to be less interested in advancing the frontiers of high communications bandwidth and more interested in consistency and quality of experience, broadly defined to include the “ilities”—reliability, manageability, configurability, predictability, and so forth—as well as non-performance-based concerns such as security and privacy. (Interest was also expressed in higher-performance, broadband last-mile access, but this is more of a deployment issue than a research problem.) Outsiders also observed that while as a group they may share some common requirements, users are very diverse—in experience, expertise, and what they wish the network could do. Also, commercial interests have given rise to more diverse roles and complex relationships that cannot be ignored when developing solutions to current and future networking problems. These considerations argue that a vision for the future Internet should be to provide users the quality of experience they seek and to accommodate a diversity of interests.

This report explores how networking research could overcome the evident obstacles to help achieve this vision for the future and otherwise better understand and improve the Internet. The report, which reflects interactions among networking researchers and outsiders (researchers from fields other than networking) at CSTB’s January 2001 workshop, as well as subsequent discussion by the Committee on Research Horizons in Networking, stresses looking beyond the current Internet and evolutionary modifications thereof and aims to stimulate fresh thinking within the networking research community. Since it is not a formal research agenda (which would, among other things, entail a much more intensive effort than is afforded by an exploratory workshop such as this), the report does not, for example, review past literature and current research programs but instead briefly characterizes past progress, current efforts, and promising directions. It focuses on three key areas in which networking research might be invigorated: measuring the Internet, modeling the Internet, and making disruptive prototypes.