MINTCar: A Tool Enabling Multiple Source Multiple Destination Network Tomography

Network Tomography

Since the last decade, network tomography for the client/server case has been widely studied (see (Castro et al, 2004) for an overview). Resulting topology is a tree where each edge represents a set of physical objects. The root is the server, leaves are clients and inner nodes are disjunction points of paths between the server and its clients. Edges can sometimes be labeled with the capacity of routers and wires belonging to the sub-path represented by it. Packet train based techniques are used to infer disjunction point for a pair of path. Probing is done for each pair of paths. Reconstruction is done most of the time using statistical techniques to estimate likelihood. Roughly speaking, it consists to collapse 2 or more disjunction points into one when capacities of the sub-paths leading to those inferred points are similar. Unfortunately, this method relies on the assertion that the resulting topology is a tree. But a tree cannot characterize the network when multiple sources and multiple destinations are involved, as stated in (Bu, Duffield, & Lo Presti, 2002). New probing techniques, reconstruction algorithms and models must then be developed for this topology discovery and performances inference.