Network function placement on virtualized cellular cores

TLDR: A linear programming (LP) formulation for the computation of NF placements that strikes a balance between optimality and time complexity is presented and results show that the LP achieves significantly better load balancing, request acceptance rate, and resource utilization compared to a greedy algorithm that performs NF placement inline with carriers' common practice today.

Abstract: The proliferation of mobiles devices, application sprawl, and the ever-increasing data volume generates significant stress on cellular networks and particularly on the cellular core, also known as the Evolved Packet Core (EPC). This is further exacerbated by the deployment of hardware appliances for the implementation of a wide range of network functions (e.g., gateways, mobility management, firewalls, network address translation), hindering any opportunity for elastic provisioning, and eventually leading to high operational costs and a significant degree of load imbalance across the EPC. Network Function Virtualization (NFV) has been seen a promising solution in order to enable elasticity in the cellular core. Applying NFV to the EPC raises the need for network function (NF) placement, which in turn entails significant challenges, due to the stringent delay budgets among cellular core components and the coexistence of communicating data and control plane elements. To address these challenges, we present a linear programming (LP) formulation for the computation of NF placements that strikes a balance between optimality and time complexity. Our evaluation results show that the LP achieves significantly better load balancing, request acceptance rate, and resource utilization compared to a greedy algorithm that performs NF placement inline with carriers' common practice today.