Scaling Telecom Core Network Functions in Public Cloud Infrastructure

Abstract

Telecommunication networks, especially 5G, typify high performance network infrastructure with very high data rates and throughput. Traditionally, high data rates meant line rate in telecom networks. As link capacities increase to 100 Gbit/s, 64byte packet rates will need to reach 150 MPPS. Technologies that attempt to deliver this high packet rate take liberties with proprietary hardware and application-enhanced networking stacks, pushing complexity and cost. Our approach challenges the need to deliver 64byte packet at line rate and instead addresses delivering good enough packet processing capacity with conventional networking stacks natively in the Public Cloud. We explore the deployment of operational telecom core network packet processing on Public Cloud infrastructure and empirically demonstrate this is in fact completely feasible. Using compute and networking available on Public Cloud as is, we realize an end-to-end operational telecom packet core delivering over 2 MPPS, ~20Gbps for 250K users, representing the network load of a Mobile Edge site. This paper re-thinks telecom core network capacity planning, addresses scale-up vs. scale out thresholds, details leveraging of Linux kernel advancements for compute and I/O efficiencies and provides a reference architecture for deployment of telecom core on Public Cloud.