Load balancing for a user-level virtualized 5G cloud-RAN

Abstract

5G cellular networks support a wide variety of applications with different Service Level Objectives (SLOs) over a shared infrastructure using virtualization. Virtualization enables network operators to allocate a tailored set of computational resources in the cloud to users from different applications based on their SLOs. Existing virtualization approaches use slices to create logically independent networks for each different application. However, these approaches fail to provide adequate performance isolation among different slices, leading to performance degradation. In this paper, we present the design and implementation of a load balancer called Dynamic Greedy Spike (DGS), for a cloud Radio Access Network (RAN) architecture with user-level virtualization. With user-level virtualization, network operators can now allocate new users to any host in the cloud irrespective of their source base station. DGS allocates these new users to different hosts to reduce interference between users and improve isolation by modeling the problem similar to weighted improper graph coloring. We implement a prototype of the user-level virtualized RAN architecture called uvRAN using OpenAirInterface. We also perform large-scale evaluations and show that uvRAN along with DGS provides significant improvement in isolation, which improves performance while reducing the compute resources required for baseband processing.