Legacy and novel network services are expected to be migrated and designed to be deployed in fully virtualized environments. Starting with 5G, NFV becomes a formally required brick in the specifications, for services integrated within the infrastructure provider networks. This evolution leads to deployment of virtual resources Virtual-Machine (VM)-based, container-based and/or server-less platforms, all calling for a deep virtualization of infrastructure components. Such a network softwarization also unleashes further logical network virtualization, easing multi-layered, multi-actor and multi-access services, so as to be able to fulfill high availability, security, privacy and resilience requirements. However, the derived increased components heterogeneity makes the detection and the characterization of anomalies difficult, hence the relationship between anomaly detection and corresponding reconfiguration of the NFV stack to mitigate anomalies. In this article we propose an unsupervised machine-learning data-driven approach based on Long-Short-Term-Memory (LSTM) autoencoders to detect and characterize anomalies in virtualized networking services. With a radiography visualization, this approach can spot and describe deviations from nominal parameter values of any virtualized network service by means of a lightweight and iterative mean-squared reconstruction error analysis of LSTM-based autoencoders. We implement and validate the proposed methodology through experimental tests on a vIMS proof-of-concept deployed using Kubernetes.

Quality of Service (QoS) management in IP networks today relies on static configuration of classes of service definitions and related forwarding priorities. Packets are actually classified according to the DiffServ architecture based on the RFC 4594, typically thanks to static configuration or filters matching packet features, at network access equipment. In this paper, we propose a dynamic classification procedure, referred to as Learning-powered DiffServ (L-DiffServ), able to detect the distinctive characteristics of traffic and to dynamically assign service classes to IP packets. The idea is to apply semi-unsupervised Machine Learning techniques, such as Linear Discriminant Analysis (LDA) and K-Means, with a proper customization to take into account the issues related to packet-level analysis, i.e. unbalanced distribution of traffic among classes and selection of proper IP header related features. The performance evaluation highlights that L-DiffServ is able to change dynamically the classification outcome, providing an higher number of classes than DiffServ. This last result represents the first step toward a more granular differentiation of IP traffic.

Network softwarization technologies challenge legacy fault management systems. Coordination and dependency among different novel software components for orchestration, switching, virtual machine and container management creates novel monitoring points, besides novel sources of faults, bugs and vulnerabilities. To cope with the high heterogeneity and granularity of software components, we propose a modular network AI framework to detect anomalies, toward closed-loop automation. We design an AI-empowered anomaly detection framework able to assess the running state and the state deviations of a softwarized infrastructure, monitored through different features grouped depending on their layer and connectcompute stack component. Our framework learns the nominal working conditions of the infrastructure, with respect to which anomalies are detected, and characterized tracing back the layer and component root cause; it includes a network state assessment technique that leverages anomalies characterization through their most visible symptoms. We implement and validate the proposed framework through experimental tests on a containerized IP Multimedia Subsystem platform.

No abstract