Designing replacement routing protocols is an important goal for continued growth of a versatile, mature Internet. Today's protocols have grown organically through interactions between router vendors and their customers. This growth has focused less on correct behavior and more on providing features that would not break existing infrastructure.
Consider the standard Internet interdomain-routing protocol, BGP. Formal guarantees of its behavior are lacking in several areas. First, no security mechanism exists to provide resilience against misconfiguration and malicious attacks. Routes are learned through update messages, but the contents of these messages are not authenticated in any way. Thus, an adversary could inject messages to divert traffic for malicious means, and erroneous messages from even a single source can have far-reaching effects. Second, BGP's convergence time is O(n!) in the worst case, where n is the number of Internet domains, and the protocol has significant transient errors, e.g., routing loops, that also increase with Internet size. Finally, BGP is not even guaranteed to converge. Routes are computed based on a composition of routing policies through the network; these policies could be based on factors as diverse as the terms of business contracts, performance issues, and security concerns in distant networks. Policies are determined and configured locally, which is extremely desirable for the Internet, given that component networks are administered by different agents. However, this expressiveness without global coordination comes at a cost: the interaction of routing policies can lead to global anomalies, such as nondeterminism and protocol oscillation.
Formal analysis of routing algorithms has shown inherent trade-offs among achieving desirable protocol design goals. However, this analysis has also yielded principles toward the design of robust protocols and has begun to formalize important desiderata. In addition, it has been shown that a combination of local and global constraints, some of which are contained in assumptions underlying the hierarchy of today's commercial Internet, can guarantee good routing behavior. However, much work is needed before realization of a robust infrastructure that does not overly limit the ability of honest network operators. Doing so not only requires additional research on the algorithms used for policy routing, but also requires accompanying tools to secure and properly configure the routing system.
Important open questions include the following:
These kinds questions require a combination of theoretical analysis, including formal methods, and experimental analysis, including large-scale deployment in realistic settings.
This workshop will bring together networking experts and algorithms experts in order to discuss recent progress and future directions in this area.