B5G: Predictive Container Auto-Scaling for Cellular Evolved Packet Core

The increase in mobile traffic which is accompanied by a random workload, variations necessitate an upgrade of mobile network infrastructure to maintain acceptable network performance. Scaling the mobile core network (Evolved Packet Core (EPC)) has attracted the attention of the research community and many scaling solutions that utilized either horizontal or vertical scaling have been proposed. Most of these solutions tend to scale the EPC entities on virtual machines (which usually takes time to instantiate) using a dedicated scaling module at the expense of an increase in overhead. In this paper, we propose a predictive horizontal auto-scaling mechanism for a container-based EPC that utilizes the embedded functionalities offered by Amazon Web Services (AWS) to scale the containerized EPC entities according to their CPU utilization. We further, formulate an optimal load balancer to distribute traffic to multiple instances to achieve fairness and maximize their throughput. The proposed auto-scaling model is implemented on the AWS cloud platform and evaluated against the number of successful attach processes, RAM, and CPU utilization. The results reveal RAM utilization does not saturate as the number of User Equipment (UE) increases for all entities and the MME CPU utilization is more affected as the number of UE's request increases.