PhD Forum: On the scalability of source routing architectures

Conventional wisdom says that a source-routed network architecture — that is, one in which packets carry explicit (partial) paths through the network, and nodes of the infrastructure simply forwards packets along those paths — is not scalable. Because of the significant benefits offered by such an architecture, its scalability merits a careful investigation. Using a combination of simulation and measurement, we propose to examine the limits of scalability of source routing architectures. We limit our focus to architectures that (i) separate routing from forwarding, (ii) separate routing from topology discovery, and (iii) use flat identifiers because these architectures are radically different than the current Internet architecture, whose scalability limits are known. We measure scalability along three dimensions: path discovery and computation; collection and maintenance of up-to-date topological information; and the latency cost of communicating with a path service.