Please note: This PhD seminar will take place in DC 1302.
Muhammad Sulaiman, PhD candidate
David R. Cheriton School of Computer Science
Supervisor: Professor Raouf Boutaba
Network slicing enables 5G and beyond networks to support diverse applications by creating logically isolated virtual networks, or network slices, tailored to their requirements. Network slices can be acquired by submitting slice requests, and require the infrastructure provider to allocate resources across the radio access, transport, and core networks. With limited infrastructure resources, maximizing long-term revenue requires jointly deciding which slice requests to admit and how to allocate their required resources across the different network segments. These decisions must be made online as requests arrive, without knowledge of future demand. A central challenge is balancing immediate revenue against the opportunity cost of occupying resources that could serve more valuable future requests. The computational complexity of joint admission and resource allocation problem together with the need for interpretable decisions further complicates this task.
In this talk, I will present an online pricing-based framework for joint slice admission control and resource allocation. The framework dynamically assigns pseudo-prices to resources that reflect their scarcity and opportunity cost, guiding resource allocation and admission decisions. First, I will introduce an exponential pricing strategy with provable worst-case performance guarantees under stated assumptions. I will then present a data-driven extension that learns its pricing parameters from historical data to improve practical performance. Evaluations on a real-world metropolitan network topology, using resource demands derived from an end-to-end 5G testbed, demonstrate mean revenue improvements of approximately 32% over a state-of-the-art deep reinforcement learning baseline and 27% over a model predictive control baseline with perfect forecasts, while reducing optimization time by roughly an order of magnitude relative to the latter.