AidOps: A Data-Driven Management of NFV configurations

NFV is expected to transform the way telecom operators deploy and manage their networks and services. Yet, network functions are usually deployed on expensive and proprietary appliances that require expertise for deployment and maintenance. In this paper we argue that technological advances in NFV need an equivalent evolution on resource-management capabilities to simplify provi- sioning while increasing operator confidence in delivering appropriate QoS at minimal costs. We propose AidOps, a framework that relies on machine learning to aid NFV orchestration leveraging workload patterns extracted from operator data. Using domain-specific knowledge, AidOps is able to optimize resource utiliza- tion according to operator constrains in terms of stability, capacity and provisioning times. We have evaluated our framework using enterprise communication apps and carrier-class services with real traffic traces.