Dynamic Edge Loading Balancing with Edge Node Activity Prediction and Accelerating the Model Convergence.

Area of Science:

• Computer Science
• Artificial Intelligence
• Network Engineering

Background:

• Mobile edge computing (MEC) networks face challenges in load balancing due to unknown edge node states and dynamic load variations.
• Effective load balancing is crucial for minimizing task processing latency in multi-user, multi-edge node environments.
• Existing methods struggle with the uncertainty and dynamics of edge node loads.

Purpose of the Study:

• To propose a deep reinforcement learning (DRL) based approach for task offloading and resource allocation in MEC networks.
• To minimize long-term average cost and balance load across edge nodes.
• To address the challenge of a priori unknown edge node load states.

Main Methods:

• Decomposition of the optimization problem into task offloading and resource allocation subproblems.
• Utilization of Karush-Kuhn-Tucker (KKT) conditions for optimal communication bandwidth and computational resource allocation.
• Application of Long Short-Term Memory (LSTM) networks for real-time edge node activity forecasting.
• Integration of deep compression techniques to accelerate model convergence.

Main Results:

• A 47% reduction in task drop rate compared to baseline schemes.
• A 14% decrease in total system cost.
• A 7.6% improvement in runtime.

Conclusions:

• The proposed DRL-based scheme effectively balances load and reduces costs in MEC networks.
• LSTM networks and deep compression enhance the efficiency and performance of the system.
• The approach offers significant improvements over existing methods for task offloading and resource allocation.