Automated Data Analytics and Resource Arbitration Scheduling for Containerized Network Functions

Abstract

The agile deployment of network functions on containers in a virtualized network infrastructure is a viable solution for realizing future diverse microservice-based applications in 5G and beyond-5G networks. Because the CPU utilization of each containerized network function (CNF) is time-varying, microservice-based applications may experience a shortage or wastage of CPU resources if a fixed amount of resources is allocated to each CNF. In this study, to realize autonomous and proactive resource control for CNFs, we proposed and implemented an automated sequential processing system that cascades CPU utilization analytics by applying least-squares support vector regression and resource arbitration scheduling for CNFs. Through experiments and numerical analyses, we prove that the proposed system is sufficiently agile to perform automated sequential processing in approximately 2 s.