Impact of Time-Varying Traffic Type on the Performance of Multilayer Networks

Traffic in backbone networks is characterized by strong seasonality, with clear patterns visible in various services and applications based on their usage throughout the day. Data-driven networks can learn these patterns to manage resources more efficiently as they become increasingly saturated. In this paper, we explore the benefits of traffic prediction and grooming across different traffic patterns. To achieve this, we simulate network operations using uniform sets of time-varying connection requests, where all demands in a simulation share the same traffic pattern related to a specific network-based service or application. Our goal is to thoroughly evaluate the robustness of the proposed techniques across diverse scenarios. The results will facilitate the design of future application-aware algorithms for the most efficient handling of each traffic pattern.