Network assurance via hybrid systems modelling and set invariance

Network Function Virtualization (NFV) along with Service Function Chaining (SFC) associate network functions with virtual resources to offer end-to-end network services. The goal is to enhance the operator’s environment with elasticity, agility, and cost-effectiveness, automating the allocation of services while guaranteeing fulfilment of their specification, i.e., achieving the Network Assurance (NA). We address the challenges of meeting Quality of Service (QoS) specifications, such as delay and security, in fluctuating traffic and a multi-tenant SFC infrastructure. First, we model the evolution of the average time delays as a piecewise affine dynamical system. By exploiting the monotonicity properties of the system dynamics, we characterise and compute the operating regions where feasible scheduling strategies guaranteeing NA exist, by calculating the maximal controlled invariant set of the system. We propose a Model Predictive Control (MPC) framework leading to recursively feasible real-time decision strategies, thus, guaranteeing NA. We present an extensive comparison with commonly used approaches from the realms of heuristics and Machine Learning and show our framework outperforms them, guaranteeing at the same time a priori all specifications.