Using network tomography for dynamic path adaptation

The paper proposes a new fast algorithm to evaluate the probability distribution function (PDF) of link delay over time by using network tomography techniques. Once we have the delay distribution, we can adapt the route of the flows between a source and its destination so that the end-to-end path is always the one that offers the least mean delay and a probability of exceeding a critical value smaller than a predefined threshold. Our mechanism, which guarantees the use of the path minimizing the end-to-end delay, is best implemented in the core network of a wireless domain. As the network tomography introduces additional flows, our mechanism is more adapted to wired peer ends, e.g. base stations, domain gateways, etc. With MPLS (multi protocol label switching) to create and maintain the routes, we validated our contribution using the Network Simulation simulator, ns2.