A Scaling Mechanism for an Evolved Packet Core Based on Network Functions Virtualization in Java

Abstract

The workload variations affect the performance of mobile networks. The scaling task is pivotal for addressing these variations. In the literature, research works have incorporated horizontal or vertical scaling in the virtualized network functions of the Evolved Packet Core (EPC) to improve its performance. However, up to now, these works exploit only horizontal or vertical scaling for achieving their aim. In this paper, we propose a scaling mechanism that utilizes horizontal and vertical scaling and considers workload variations for improving performance in EPC. This mechanism is threshold-based, straightforward, and implementable in real LTE-EPC scenarios. We also develop a mechanism prototype and deploy it in a real public cloud. In this cloud, we conduct a prototype evaluation, regarding registrations per second, latency, CPU, and RAM, and considering a varying workload. The evaluation results reveal that our mechanism increases the registrations per second about 308% and decreases the corresponding latency approximately 70% regarding an EPC without scaling while keeping the CPU usage lower than 90% and the used capacity of registrations per second between 65% and 90%. These results corroborate the importance of used both horizontal and vertical scaling to improve EPC performance, handle workload variations, and save resources.