Predictive Autoscaling Orchestration for Cloud-native Telecom Microservices

Abstract

Mobile traffic is dramatically increasing in recent several years with the evolution of mobile network toward 5G era. Network function virtualization (NFV) and cloud computing provide more flexibility and elasticity for mobile networks. The autoscaling orchestrator allows adjusting the number of virtual network functions (VNFs) in a telecom cloud platform corresponding the devices demand (smartphone, IoT, robots,...). However, the adjusting process produces the delay resulting on instance creating to make services available with the enough resource. The predictive mechanisms with workload forecasting enable the improvement in performance of the autoscaling orchestration system. In this paper, we investigate predictive autoscaling in the orchestration system for virtualized mobile networks. We propose a cloud-native approach for stateless telecom services. We also consider and investigate realtime prediction to detect the peak load or burst request of resource. We develop longterm forecasting for periodical resource that detect seasonal workload demand and provide the resource plan to cloud provider. Finally, we evaluate the accuracy of these predictive mechanisms with a growing workload with several accuracy criteria. We develop our telecom testbed using containerization technology and these approaches are integrated with Kubernetes orchestrator.