An accelerated Benders decomposition approach for virtual base station formation in stochastic Cloud-RANs

Abstract

In Cloud-RANs, the baseband unit (BBU) is separated from the remote radio head (RRH) at the cell site and migrated to a server cloud to decrease deployment and maintenance costs. A base station (BS) can be formed as a network slice in which the virtual network functions (VNFs) of its BBU can be placed on any server by network virtualization (NFV) technology and logically connected by software-defined networking (SDN). However, the uncertain demands of users make it very difficult to determine a network slice with appropriate VNF placement and flow routing between VNFs. We aim to minimize the cost of providing BS services while meeting their time-varying demands. We consider the network slice scheme and fronthaul establishment jointly and formulate this as an integer linear programming (ILP) problem. We propose scenario classification Benders decomposition (CBD) based on Benders decomposition (BD) but with fewer Benders cuts, accelerating the solution convergence. Moreover, we propose an efficient heuristic approach to obtain good solutions for the ILP problem based on the solutions of its relaxed problems obtained by BD or CBD. The numerical experiments show that when combining our heuristic approach with CBD, the results are close to the optimal solutions of the ILP problem, while the computation time is significantly shorter than those of previous approaches.