ADMIRE: collaborative data-driven and model-driven intelligent routing engine for traffic grooming in multi-layer X-Haul networks

X-Haul aims to provide a unified transport network for integrating mobile fronthaul/midhaul/backhaul. Its architecture is inherited from the IP-over-WDM paradigm, in which the upper layer is an electrical packet-switched network and the bottom layer is an optical circuit-switched network. Traffic grooming in X-Haul poses a big challenge, as the mobile traffic load has shown. As a multi-layer routing problem, traffic grooming has been well studied through some classic mathematical models, such as the auxiliary graph (AG). Recently, a new approach based on machine learning has been attracting much attention for its network routing decision making. According to previous studies, either of the two approaches has its limitations, but the combination of the two may achieve profit maximization. To this end, we propose and demonstrate, for the first time, to our knowledge, a collaborative data-driven (using machine learning) and model-driven (using experiential knowledge) intelligent routing engine (ADMIRE) for traffic grooming in an X-Haul testbed with a real mobile dataset. The principle of ADMIRE is to exploit the capability of machine learning for an accurate AG model in a dynamic network environment. We compare ADMIRE with a traditional model-driven approach (i.e., the AG), and the evaluation results show that ADMIRE can achieve good performance and a strong generalization ability. In addition, we also verify the influence of data correlation on ADMIRE.