Model-based diagnosis techniques for Internet delay diagnosis with dynamic routing

In this paper we propose a model-based approach to diagnose latencies in computer networks. We formalize this problem as a model-based diagnosis (MBD) problem and propose a range of methods to solve it. Three solution approaches are proposed: a conflict-directed approach, a constraint satisfaction approach and a linear programming approach. We discuss the pros and cons of these approaches and describe which approaches are suited to handle which network routing policies. In particular we handle this work networks with static routing policies, where there exists a static route between every pair of end-users, as well as two common types of dynamic routing policies, where information between a pair of end users may pass via more than a single route. The performance of the proposed approaches is demonstrated experimentally on two domains: the standard NS2 network simulator and on parts of the Internet topology obtained from the Route Views project. Both able to find diagnoses fast for network models with 1,000 nodes.