ECODE Project Overview
Objectives
- Design and experiment a cognitive routing system that aims at addressing future Internet challenges:
- Security and diagnosability: monitor the path performance by combining adaptive passive and active measurements, and cooperatively detect traffic anomalies (leading to performance decrease) so as to detect intrusions and attacks;
- Availability/resiliency and accountability: informed path ranking based on metrics, path re-routing to other links in cases of failure, and traffic flows correlation by routers to diagnose and predict performance deviation over time (with respect to profiles), and adapt these profiles so as to maintain an acceptable resource usage;
- Scalability and quality of the routing system: by detecting events that are detrimental to the routing system dynamics (convergence, stability/ robustness, and stretch), decide and efficiently react to such events.
This cognitive routing system combines novel networking techniques with on-line, adaptive and distributed machine learning techniques. The resulting system intends to preserve as much as possible original Internet design principles (end-to-end, transparency, etc.).
Prototype cognitive routing system on XORP open routing emulation platform and validate the machine learning techniques on physical (iLAB) and virtual (e.g., OneLab) experimental facilities.
Methodology
- Our methodology relies on cross-fertilization between the networking and machine-based domains to form a cognitive routing system answering the operational and new Internet challenges. Indeed, they are similar in nature to the conditions traditionally encountered in classical machine learning problems:
- Nature: the events cannot be well characterized even when examples of such an event are available (inherent complexity in precisely characterizing an event);
- Relationship: the correlations and trends between events are hidden within large amounts of data that are associated to these events;
- Environment: the conditions are changing over time (this is particularly the case for the routing environment but also variability of user demands, expectations and behaviours);
- Quantity: the amount of available data is too large for handling by human intervention;
- Evolutive: new events are constantly detected/discovered.