An algorithm for routing optimization in DiffServ-aware MPLS networks

This paper addresses the constrained-based routing problem in DiffServaware MPLS networks. We consider a dynamic context in which new requests appear over time, asking for reconfigurations of the previous allocation. In the classical approach, a multi-phase heuristic procedure is adopted: the new requests are evaluated considering available bandwidth; if the bandwidth is not sufficient, preemption and rerouting of one or more connections are performed in sequence. As an alternative, we propose a single-phase approach that simultaneously takes into account both preemption and rerouting. In contrast with the standard approach, we always find an optimal solution when enough bandwidth is available. Otherwise, we apply a heuristic post-processing procedure in order to minimize unsatisfied commodities. The routing problem is modeled as a Multicommodity Flow Problem (MCFP) with side constraints, which is solved by a Column Generation approach. Namely, a sequence of restricted MCFPs is solved, by including new routing paths only if necessary. When new requests are routed, the use of existing paths is preferred in order to reduce preemption. Computational experience on real networks shows that the overall approach is able to (i) obtain a good exploitation of the network resources, (ii) achieve a remarkable acceptance rate and, (iii) hold down the impact of rerouting.