Collaborators: François Baccelli (INRIA) Paul Tune (UoA) Jean Bolot (Technicolor) Renata Teixiera (UPMC) Timur Friedmann (UPMC)
Internet Measurement is a rich field devoted to the development of accurate experimental, measurement, analysis and inference techniques to shine light into the vast unknown that is `what is going on in the Internet'.
This project pursues several strands, including:
Active probing: this is the practice of sending streams of `probe' packets into the network, to infer network behaviour through observing their end-to-end experience (if they are lost, delayed). The current focus is to explore the `convex network' approach to solving problems in optimal probing (getting the most information per-probe).
Traffic Sampling: in high speed environments like inside routers in the Internet core there are far too many packets to measure them all. We research optimal sampling, sketching, and `skampling' techniques to measure metrics like the flow-size distribution at ultra high-speed with minimum variance.
Optimal Traceroute: Traceroute is a well-known tool for measurement of the paths that packets take when traversing the Internet. The project refines its operation to provide statistical guarantees when used for topology discovery.
Network Anomaly detection: what is normal traffic, and what is an attack? This is the core, difficult question at the heart of much of network security. This project explores both statistical and machine learning approaches to this problem, and well as non-linear filtering.
