Enable optimal resource utilization for VNF through Intelligent Cloud Platform

Abstract

Network Function Virtualization (NFV) is a network architecture built to run on commercial off-the-shelf (COTS) servers in a virtualized environment. Lower layers network function of telecom network elements like PHY, MAC and Bearer are hosted as Virtualized Network Function (VNF). NFV Management and Orchestration (MANO) referred as cloud platform allocates, monitors and controls the cloud-computing resources. In the Telecom world, resources are statically provisioned so that the resource utilization is optimal in order to meet the KPIs. Owing to this resource dimensioning and to secure a stable deployment cost, customers require a static configuration of cloud system resources like CPU, memory and network for VNFs. When these cloud resources are over-provisioned in view of higher resource utilization, the VNF instances will have workloads contributing to active and lean periods based on user applications requirement (throughput, latency) in the network. In a cloud environment, ad-hoc workloads, like performance testing, monitoring(vProbe), diagnostics etc., need to be deployed occasionally. In this paper, we propose an intelligent cloud platform that can enable an operator to exploit the static resources which are under-utilized. We experimented with Auto-Regressive Integrated Moving Average (ARIMA) model and our results show that the proposed intelligent cloud platform is able to forecast CPU, memory, and network utilization with an accuracy of 95.4%, 93.7% and 96.2% respectively. Further, VNFs are able to run additional workload during lean period with average CPU, memory, and network utilization improved by 38%, 19% and 20% respectively.