Architectures de transport et de services multi-fournisseurs

The dissertation presents various technical solutions to improve the level of collaboration among providers in support of inter-provider network services. The scientific contribution embraces different networking research facets, from IP routing to G-MPLS provisioning and network design optimization, applying concepts from graph theory, game theory and operations research. By an in-depth analyzis of recent Internet routing traces, we show that the current inter-domain (connection-less) routing suffers from a lack of coordination that produces inefficiencies and frequent route deviations. With respect to this issue, relying on concepts of non-cooperative game theory, we propose coordination strategies to improve the current BGP routing across peering settlements, while preserving the providers' independence and respective interests. We show that their implementation can avoid congestion on peering links, reduce significantly the routing cost and successfully control the route deviations. The mathematical model can be extended to support a new form of peering agreement extended to multiple providers, but its adoption may appear too weak with respect to alternative solutions able to guarantee end-to-end cross-provider Quality of Service (QoS). The support of strict end-to-end QoS constraints for added-value services imposes, indeed, a higher level of collaboration on the multi-provider agreement. It is required to reserve resources for own services in other providers' networks. These requirements bring towards a new interconnection model, the \lq \lq provider alliance\rq \rq, as a cooperative framework that providers shall deploy to allow dynamic connection-oriented service routing and provisioning. We define the functional architecture of a service plane managing service-related data within the provider alliance, together with the instantiation and activation of multi-provider tunnel and circuit services. We highlight the required protocol extensions for the distributed (router-level) path computation and the dynamic resource reservation, which have been implemented and validated in a testbed. We define, moreover, specific AS-level routing algorithms that scale with the proposed model, supporting pre-computation and directional transit metrics. Finally, we show how providers shall cooperate also to statically reserve link resources, in an optimal and distributed fashion, modelling the economical incentives and the strategic position of each provider in such a cooperation with the application of concepts from cooperative game theory (precisely, the Shapley Value concept). In the second part of the dissertation, we tackle more physical issues related to the provisioning of tunnels and circuits across Internet eXchange Point (IXP) infrastructures. We present a novel very-high-capacity optical transport architecture, called the Petaweb, as a possible next generation IXP solution and, more generally, as a possible very-high-capacity transport architecture. It consists in a regular direct interconnection scheme of electronic access nodes via optical switches disconnected from each other. This structure can allow a simple inter-provider G-MPLS signalling, can drastically simplify traffic engineering operations, and can facilitate modular upgrades of network elements, at the expense of potentially higher installation costs. We formulate the design dimensioning problem of the Petaweb composite-star topology, which is NP-Hard, and propose a scalable and efficient heuristic approach. Moreover, we propose a quasi-regular structure for the same transport architecture, less costly and slightly more complex (requiring wavelength conversion), for which we also formulate the design problem and propose an efficient heuristic. We argue by simulations that the physical dimensioning of classical multi-hop optical networks under additive path metric minimisation (such as the delay) would produce a solution that tends toward a quasi-regular Petaweb structure. To conclude, we analyze how practically a network planner decision-maker shall trade-off -- when discriminating among many Petaweb solution alternatives -- the various performance criteria with the level of reliability, survivability and availability