Autoscaling and resource placement in containerized based NFV

Abstract

Current telecommunication networks are plagued with ever increasing demands of ser vice agility and scalability. Network function virtualization (NFV) and Software defined Networking (SDN) are paradigms which addresses the issue of slow scaling and reduced time to introduce new services, which are hindered in network services built up on proprietary, dedicated hardware. However, to maximize the advantages of the above mentioned paradigms, efficient scaling approaches have to be researched to efficiently manage resources in a containerized based NFV system. In this thesis we propose a ma chine learning algorithm based on multi-input and multi-step Long Term Short Memory (LSTM) to predict traffic throughput demands. Furthermore, we propose a cloud native implementation and evaluate the performance of the autoscaling algorithm using actual data in a virtualized mobile core network.