Related Experiment Video
Updated: Jul 2, 2025
05:30
Large Scale Energy Efficient Sensor Network Routing Using a Quantum Processor Unit
Published on: September 8, 2023
551
A Temporal Deep Q Learning for Optimal Load Balancing in Software-Defined Networks
Aakanksha Sharma1, Venki Balasubramanian2, Joarder Kamruzzaman2
1Melbourne Institute of Technology (MIT), Melbourne, VIC 3000, Australia.
Sensors (Basel, Switzerland)
|February 24, 2024
Summary
Temporal Deep Q learning (tDQN) optimizes Software-Defined Networks (SDNs) for Internet of Things (IoT) traffic. This intelligent approach minimizes network latency and improves performance without adding controllers.
Area of Science:
- Computer Science
- Network Engineering
- Artificial Intelligence
Background:
- The Internet of Things (IoT) generates massive network traffic, overwhelming traditional networks.
- Software-Defined Networks (SDNs) offer better traffic management but face control plane bottlenecks and load balancing issues in multi-controller setups.
- Existing dynamic controller mapping solutions struggle with controller placement and real-time load balancing for fluctuating traffic.
Purpose of the Study:
- To propose an intelligent solution for dynamic controller mapping in multi-controller SDNs to minimize network latency.
- To address the challenges of unpredictable IoT traffic growth and heterogeneous device complexity.
- To enhance the performance of dynamic SDN (dSDN) using a self-learning reinforcement-based model.
Main Methods:
- Implementation of Temporal Deep Q learning Network (tDQN) within the dSDN controller architecture.
- Utilizing a reinforcement learning agent with a reward-punish scheme for optimizing switch-controller mapping decisions.
- Formulating a multi-objective optimization problem to dynamically divert traffic based on flow fluctuations.
Main Results:
- The tDQN approach effectively optimizes dynamic flow mapping and reduces network latency without increasing the number of controllers.
- Extensive simulations demonstrate superior performance compared to traditional networks, eSDN, and dSDN.
- Significant improvements observed in throughput, delay, jitter, packet delivery ratio, and packet loss across varied network scenarios.
Conclusions:
- Temporal Deep Q learning provides an intelligent and effective solution for managing complex and dynamic SDN environments, particularly for IoT.
- The tDQN model successfully balances load and minimizes latency, outperforming previous methods.
- This research offers a scalable and efficient approach to future-proof SDN infrastructure against exponential traffic growth.
