Conference Paper

The combination of traditional optical transport network paradigms with the advent and necessities of edge computing opens up new challenges for optimizing both the optical layer and the data-center (DC) infrastructure. In the specific case of ring/horseshoe metro access topologies, the flexibility to divert services to a specific DC with low latency is highly dependent on the node architecture that is employed. Consequently, the optimal trade-off between consolidating services in larger DCs and the number of transponders required for transporting traffic will shift depending on the possibilities enabled by the network itself. Furthermore, traditional north/south traffic to/from the core is still expected to be significant in these topologies, providing opportunities to leverage existing optical bandwidth for improved cost effectiveness. This process requires a careful dimensioning in order to optimally consider all impacting factors and select which services to assign where. We model this problem with a mixed integer quadratically constrained problem (MIQCP) that optimizes DC resources for a given availability level according to services' mean and peak loads. This optimization accounts for the interdependencies with the required optical transponder costs and the constraints imposed by service latency, as well as the specific restrictions imposed by optical nodes based on fixed/reconfigurable add/drops, or filterless solutions. The resulting analysis not only provides optimal solutions to specific network/DC planning instances, but also identifies general deployment guidelines for metro access networks in terms of desired node architectures for each network scenario.