Optimal monitoring in large networks by Successive c-optimal Designs

We address the problem of optimizing the use of Network monitoring tools, such as Netflow, on a large IP network. We formulate a convex optimization problem which allows one to handle, in a unified framework, the combinatorial problem of selecting the “best” set of interfaces on which Netflow should be activated, and the problem of finding the optimal sampling rates of the network-monitoring tool on these interfaces, when the aim is to infer the traffic on each internal Origin-Destination (OD) pair. We develop a new method, called “Successive c-optimal Design”, which is much faster than the classical ones. It reduces to solving a stochastic sequence of Second Order Cone Programs. We give experimental results relying on real data from a commercial network, which show that our approach can be used to solve instances that were previously intractable, and we compare our method with previously proposed ones.