Page 346 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

• SDN Controller: A distributed SDN architecture dynamically controls switch and base
station states based on optimization outputs.
• Simulation & Evaluation: The system will be simulated (e.g., using Mininet), evaluating
energy savings, throughput, and link utilization.

Topology Control: The network topology is dynamically adjusted per time slot based on
predicted traffic demands. The SDN controller evaluates current link utilization and forecasts,
determining which links to activate or put into sleep mode. This uses a real-time link activation
matrix that maps traffic load thresholds to energy-optimal link states. Only the minimum number
of active links needed for traffic flow and redundancy are kept operational to conserve energy.

Max-Min Routing: Routing is framed as a multi-objective optimization problem that aims to
maximize the minimum available bandwidth (max-min fairness) and minimize the energy
consumption of active links. This is solved using Integer Linear Programming (ILP) with
constraints on link capacities, per-link energy costs, and flow conservation. The optimization is
performed periodically (e.g., hourly), with results sent to the SDN controller to update routing
and link states. For large-scale settings, a GNN + DRL heuristic is used to learn efficient routing
strategies through simulation and real-time feedback.

Path Selection with GNN + DRL:

Traditional methods like LP or weighted Dijkstra are limited in dynamic, large-scale environments.
We propose a hybrid ML approach using Graph Neural Networks (GNNs) to represent the
network topology and Deep Reinforcement Learning (DRL) to learn adaptive routing policies.
This enables real-time decisions that optimize both traffic handling and energy use. Prior
studies show that GNN + Distributed RL outperform classical methods in domains like V2X
and 5G backhaul [5][6]. To ensure both high throughput and energy efficiency, we combine
traditional ILP-based optimization with learning-based GNN + DRL models, forming a scalable
and adaptive multi-objective routing solution.

Hierarchical SDN Controller Design:

A two-tier SDN architecture is used. The Global Controller coordinates inter-regional policies
and traffic flow. Regional/Local Controllers manage smaller domains like programming
data-plane switches and controlling base stations. This structure scales well for dense 5G
deployments, supporting quick local responses with centralized oversight.

Use Case Status: Initial Prototype

Partners : No external partners are currently involved. All R&D is conducted within the 5G Use
Case Lab, IIT Gandhinagar. Future collaborations with industry stakeholders and government
bodies (e.g., DoT India, BSNL) may be considered.

2�2 Benefits of the use case

This solution directly addresses energy efficiency in 5G networks by optimizing resource
allocation and dynamically adjusting network topology based on real-time traffic demands.
Given the high electricity consumption within the telecom industry, this approach helps
minimize energy wastage and contributes to more sustainable network operations.

The system also promotes innovation in network infrastructure by integrating Software-Defined
Networking (SDN) and AI-driven optimization. This enables energy-efficient planning and

310