Source routing in networks with uncertainty: inference, sensitivity and path caching

In this paper we study source routing in an environment where imperfect state information is the norm. The uncertainty involved in several aspects of the routing process renders the route choices less than “optimal”. We start by conducting an experiment that compares the performance of an “inference”-based routing scheme to that of the traditional approach based on delayed link state broadcast. We then resort to a set of simple models to investigate to what extent the “crude” routing decisions based on limited statistical information conform to the ideal choices. In the conventional routing context, we identify a useful measure, the gap, which quantifies how successful a “crude” routing decision is likely to be. In the quality of service routing context we explore the possibility that a route choice based on limited statistical information is the “most likely” path to satisfy the user requirement. We also discuss the role of critical points, whose relative position affects the robustness of the routing decisions with respect to uncertain user requirement. Simulations establish the existence of gap and critical point in a realistic setup. The impacts of these observations on the effectiveness of a simple path caching scheme are then discussed.