Complexity-Performance Trade-offs in Robust Access Point Clustering for Edge Computing

Edge computing penetration in mobile access networks is the next barrier to break in communication networks. The virtualization of radio access functions currently under study is expected to trigger the deployment of edge cloud facilities in telecom operator points-of-presence and central offices, to serve the virtualization of both application servers and network functions. The problem of clustering network access points for their assignment to edge cloud facilities has been addressed in the literature. Nonetheless, the inclusion of key-performance indicators such as robustness against traffic variations in the optimization process can increase its complexity excessively while hindering the achievable performance. Leveraging on previous work in this area, in this paper we explore how to reduce time and spatial complexity while introducing additional a robust access point assignment target by using a spatial clustering pre-processing in the optimization problem, grouping together access points based on their spatio-temporal traffic profile. By extensive simulation against real traffic traces and network maps, we show under which conditions we can outperform existing methods at the state of the art. The obtained results show that our approach helps reducing time and space complexity for small to medium instances, indicating the geographical scale at which these operations could be run in a near-real-time manner.