A Scalable Heuristic For Hybrid Igp/Mpls Traffic Engineering - Case Study On An Operational Network

In current IP networks, a classical way to achieve traffic engineering is to optimise the link metrics. This operation cannot be done too often and can affect the route of a lot of traffic. Multiprotocol Label Switching (MPLS) opens new possibilities to address the limitations of IP systems concerning traffic engineering thanks to explicit label-switched paths (LSPs). This paper proposes a new method based on simulated annealing meta-heuristic to compute a set of LSPs that optimise a given operational objective. The hybrid IGP/MPLS approach takes advantage of both IP and MPLS technologies and provides a flexible method to traffic engineer a network on a day to day basis. We illustrate the capabilities of our method with some simulations and a comparison with other techniques on an existing operational network. The results obtained by setting up a small number of LSPs are nearly optimal and better than by engineering the IGP weights. Moreover, although it could be combined with a static setting of the latter, SAMTE alone gives already the same results as this combination in much less CPU time, which thus allows an administrator to keep its initial and meaningful IGP metrics in his network.