CAREER: Dependable Network Communication

University Of Washington, Seattle WA

Investigators

Abstract

The researcher's agenda over the next decade is to improve the robustness of network communication services to the point that they are suitable for applications that depend on assured connectivity. In this proposal, the researcher focuses on the key goal of containing the effects of router misconfigurations, implementation bugs in routing protocols, and other insider faults that, if left unchecked, can cause widespread loss of connectivity. Dependable communications are of clear importance as critical applications in the areas of aviation, medical services, emergency services, utilities and defense become integrated with the Internet. Yet the Internet today cannot be depended on because the routing protocols that provide connectivity are themselves fragile. They take a bimodal approach to security: well-developed cryptographic techniques are used to authenticate trusted entities and protect the protocol from untrusted attackers, but no further checks are placed on the information provided by an entity once it has been authenticated. The result is that once an inadvertent error or attacker slips in, the scope of error is potentially unbounded. As one example of a well-known, spectacular failure, misconfiguration at a Virginia-based ISP caused most Internet backbone traffic to be misdirected for up to two hours in April 1997. Data from one study of Internet failures suggests that insider faults account for roughly five times more trouble tickets than malicious attacks, such as denial-of-service. The researcher proposes to work towards the design of routing protocols that are able to efficiently tolerate the above kind of insider faults. Detecting and containing these faults is a challenging problem because traditional security techniques are often ineffective. For example, authentication can validate what entity sent which message, but not that the entity is behaving correctly. The key to his approach is to extend routing protocols with information that can be used by the participants to consistency-check the behavior of each other. This is a novel strategy that differs from most of the prior work, which is focused on adding security in the context of existing routing protocols. The researcher illustrates the approach in this proposal by describing his research on a robust congestion signaling protocol, where it was applied to substantial advantage in a related domain. To begin this work, the researcher will conduct a measurement study of configuration errors in BGP, the routing protocol used across the backbone of the Internet. I have already started this task, and include some preliminary results in the proposal. Such a study is important because there is little data to quantify the kind, prevalence or impact of insider faults. Armed with these results, the researcher will design routing protocols that limit the loss of connectivity caused by common faults. the researchers philosophy is to first put aside deployment considerations to focus on what can be achieved as a more fundamental result with a clean-slate design, and then map the designs into the context of existing routing protocols. Specifically, he will reason about the minimal mechanism required to handle different kinds of faults, and evaluate the costs of that mechanism by using a combination of implementation, simulation, and comparison to alternatives in the literature. The researcher's approach is also to tackle the simplest, non-malicious insider faults first and work towards progressively more complex classes of faults, rather than beginning with the design of a .Byzantine robust. protocol. This has the advantages of both breaking a known, hard problem into pieces, and exposing the increased computational costs of tolerating more complex faults. At all of the above stages, the researcher will cross-fertilize my research and education activities as described in the proposal. The researcher will bring his research into the classroom to enliven lectures, and bring students, their projects, and overlapping infrastructure such as the proposed animations back into his research. If successful, this work will deepen the understanding of dependable network communication and how routing protocols can efficiently contain faults. This in turn will lay a foundation for research and education on dependable distributed systems that rely on the composition of routing protocols with other components such as transport protocols and name resolution.