We study the problem of Distributed Unit (DU) & Centralized Unit (CU) placement in Open RAN for reducing the energy footprint of the network, under explicit distance & bandwidth constraints on real-world Radio Unit (RU) topologies. We show that the DU / CU placement can be formulated as a minimum dominating set (MDS) problem on graphs derived from latency & bandwidth constraints, enabling exact solutions that minimize the number of deployed nodes. To further refine the placement while preserving this minimum deployment cardinality, we propose a sequential distance-weighted MDS approach that selects, among all minimum-cardinality solutions, the one reducing the load transport cost. We evaluate the proposed method on a real-world node topology using a detailed energy model capturing both computational and transport costs. The results show that the MDS formulation significantly reduces infrastructure footprint compared to a clustering-based baseline, leading to a global RAN energy gain of around 14%, while the sequential refinement provides additional gains reducing latency and transport energy cost.
Hiba Hojeij, A. Aravanis, Sahar Hoteit et al.· International Mediterranean...· 0 citations
With the emergence of autonomous vehicles and the ever-increasing volume of generated data, the Mobile Edge Computing paradigm has been proposed to address challenges related to latency and computational capacity. However, static vehicle-to-MEC association policies fail to meet these requirements due to the highly dynamic nature of vehicular networks. To address these challenges, this PhD research focuses on mobilityaware orchestration in MEC-enabled vehicular networks. In our first contribution, we introduce an ETSI-compliant proactive migration framework based on proximity-triggered migration notifications and a mobility-aware task migration strategy to relocate vehicular applications during mobility. The second contribution extends this work toward adaptive orchestration by formulating allocation and migration as a decision-making problem and developing a learning-based framework with a Simu5G–Python interface and a fairness- and delay-aware Maskable PPO agent. Results show improved response time, lower deadline miss rate, and balanced resource utilization under dense vehicular conditions. Our current work focuses on explainability methods to understand the learned policy and support trustworthy deployment while improving performance in terms of E2E delay, deadline miss rate, and fairness in resource utilization.
Rim Sayegh, Hela Marouane, Sahar Hoteit et al.· IEEE Conference on Network S...· 0 citations