Why (and How) Networks Should Run Themselves

Abstract

The proliferation of networked devices, systems, and applications that we depend on every day makes managing networks more important than ever. The increasing security, availability, and performance demands of these applications suggest that these increasingly difficult network management problems be solved in real time, across a complex web of interacting protocols and systems. Alas, just as the importance of network management has increased, the network has grown so complex that it is seemingly unmanageable. In this new era, network management requires a fundamentally new approach. Instead of optimizations based on closed-form analysis of individual protocols, network operators need datadriven, machine-learning-based models of end-to-end and application performance based on high-level policy goals and a holistic view of the underlying components. Instead of anomaly detection algorithms that operate on offline analysis of network traces, operators need classification and detection algorithms that can make real-time, closed-loop decisions. Networks should learn to drive themselves. This paper explores this concept, discussing how we might attain this ambitious goal by more closely coupling measurement with real-time control and by relying on learning for inference and prediction about a networked application or system, as opposed to closed-form analysis of individual protocols.

Distributed Consistent Network Updates in SDNs: Local Verification for Global Guarantees

“…In particular, it will also be interesting to consider the use of randomized [21] and approximate [22] solutions to improve our approach, provide extensions to further consistency properties such as waypoints [23] and congestion [24], as well as seamless updates [25], but also the inherent connections to self-stabilization [26]. More generally, we believe that our approach can provide interesting new perspectives on emerging self-driving networks [27], which center around fine-grained and fast adaptions of networks reacting to their environment, and may hence benefit from our distributed approaches. Furthermore, it will be interesting to investigate opportunities coming from emerging programmable dataplanes, to speed up our approach further, as well as to generalize it to additional use cases.…”

The Adversarial Machine Learning Conundrum: Can the Insecurity of ML Become the Achilles' Heel of Cognitive Networks?

“…Multiple explanations for the sensitivity of the ML models to adversarial examples have been provided in the literature [4] including the nonlinearity of the DNN models, which can assign random labels in areas that are underexplored in the training set. But such a hypothesis fails to explain the transferability 1 of adversarial examples from one ML model to another. In addition, it is not only the nonlinear DNN models that suffer from these attacks, but linear models have also been shown to be vulnerable to adversarial examples [4].…”

“…Real-time network telemetry consists of supervised ML and feature engineering but there are more complex tasks in the self-driving cognitive networks (i.e., data-driven analysis and decision making). In order to perform these tasks, the network must have the ability to interact and adapt according to network conditions [1]. Deep reinforcement learning (DRL) provides the ability to interact, learn, and adapt to the ever-changing network conditions and it is expected to be heavily utilized in future self-driving cognitive networks.…”

“…The cognitive networking idea-a recurring motif in networking research that has been expressed in various guises such as autonomic networking, self-organized networking, knowledge-based networking, and most recently as self-driving networking [1]. Now, this idea appears to be within grasp thanks to the tremendous strides made in the field machine learning (ML) that have transformed the entire fields of vision, language, speech, and information processing.…”

Adversarial Machine Learning Attack on Modulation Classification

“…The success of machine learning (ML) in computer vision and speech processing has motivated the networking community to consider deploying ML for the automation of networking operations. Recently, new networking paradigms like cognitive self-driving networks [5] and most recently knowledge defined networking [10] have also emerged that depend on and facilitate the extensive utilization of ML schemes for conducting networking tasks. Recently ML has successfully applied on different cognitive self-driving networking tasks such as modulation classification [14] and representation learning of radio signals [12].…”