Amaury Van Bemten
Lehrstuhl für Kommunikationsnetze (Prof. Kellerer)
Amaury Van Bemten studierte Ingenieurinformatik an der Université de Liège (Belgien) und schloss sein Studium im Juni 2015 mit dem Grad MSc of Computer Science and Engineering ab. Im September 2015 ist Amaury dem Lehrstuhl für Kommunikationsnetze der Technischen Universität München beigetreten. Sein Forschungsschwerpunkt liegt auf der Ressourcenoptimierung für vorhersehbarem Latency für Kommunikationsnetzen.
Design, Implementation and Evaluation of Solutions for Predictable Latency in Industrial and Cloud Networks.
Operators are more and more looking to provide predictable performance to their tenants and/or applications. Besides being a key performance metric, latency is also the most complicated and expensive to guarantee, as networks usually leverage the multiplexing gain to increase utilization, an approach which does not match with worst-case latency computations.
Our work aims at designing, implementing and evaluating a system able to provide latency guarantees to tenants in a cloud network or to applications in an industrial network while avoiding oversubscription and trying to maximize the utilization of the network.
Such as system based on a centralized controller is mostly based on routing algorithms for the determination of routes through the network and on network models for the quantification of latency and potentially other QoS metrics.
Routing algorithms and protocols have been thoroughly investigated in the literature. However, because of the distributed nature of legacy networks, only partial or short-term knowledge could be assumed. The central and complete knowledge offered by SDN allows the definition of complex centralized routing algorithms for solving more complex problems. Our work mostly focuses on routing algorithms for constrained shortest path (CSP) routing, routing through service function chains (SFCs) and their interplay with a system for the provisioning of predictable latency.
Most of our evaluation results are gathered on an interactive web interface: LORA.
In order for a routing algorithm to provide paths with the desired QoS, a model has to provide it with the relevant parameters by abstracting them from the state of the network. For hard latency guarantees, network calculus, because of its ability to compute deterministic performance bounds, positions itself as a promising modeling tool. Our work deals with the definition of network models based on network calculus that can be used as part of a complete system providing predictable latency to tenants in the cloud or to applications in an industrial environment.
Do not hesitate to contact me if you are interested!