Contributions to Traffic Engineering and Resilience in Computer Networks

The Internet traffic is constantly increasing following the emergence of new networkapplications like social networks, peer-to-peer, IP phone or IP television. In addition,these new applications request better path availability and path quality.Indeed the efficiency of these applications is strongly related to the quality of the underlying network.In that context network operators make use of traffic engineering techniquesin order to improve the quality of the routes inside their network, but alsoto reduce the network cost of increased traffic handling with a better utilization of existing resources.This PhD thesis covers several topics of Traffic Engineering andFast Restoration in IP/MPLS networks.Our first contribution is related to the definition of a well-engineered network. In the literaturemathematical formulation of Traffic Engineering (TE) requirements are very diverse.We have thus performed a comparative study of many objective functions, in order to differentiate them andchoose in a rational way the one that best reflects Traffic Engineering goals.We have also designed a method approaching optimal TE, whereby we dividethe traffic matrix in N sub-matrices and route them independently, based on the derivatives of the objective function. The second topic addressed in this work concerns link weight optimizers (LWOs).Link weight optimization is the traffic engineering {it « standard »} technique in networks runninglink state routing protocols (which are widely used in transit networks).These link weight optimizers suffer from several limitations due to the BGP (Border Gateway Protocol) Hot-Potatorule, which is basically not considered by such optimizers.Therefore we have proposed a BGP-aware link weight optimization method that takes problematic Hot-Potato effectsinto account, and even turns them into an advantage. We have also studied how LWOsbehave in big networks which have to use BGP route reflectors. Finally we have studiedwhether forwarding loops can appear or not when traffic is split among multipleequivalent egress routers, an optional BGP feature that we did use in our Hot-Potato aware LWO.Our last contribution concerns network resilience. We have proposed a solution for a rapid recovery from a link or node failurein an MPLS network. Our solution allows a decentralized deployment combined with a minimal bandwidth usage while requiring onlyreduced amount of information to flood in the network. Thismethod is the first that makes possible a decentralized deployment combinedwith an optimal resource consumption.To easily simulate and test the methods proposed in this work, we have also contributed to the development of TOTEM - aTOolbox for Traffic Engineering Methods.