Structure-Aware Reinforcement Learning for Node-Overload Protection in Mobile Edge Computing

Mobile Edge Computing (MEC) involves placing computational capability and applications at the edge of network, providing benefits such as reduced latency, network congestion, improved performance applications. The reliability MEC degrades significantly when server(s) in cluster are overloaded. In this work, an adaptive admission control policy to prevent node from getting overloaded is presented. This approach based on a recently-proposed low complexity RL (Reinforcement Learning) algorithm called SALMUT (Structure-Aware Learning for Multiple Thresholds), which exploits structure optimal multi-class queues average-cost setting. We extend framework work overload-protection problem discounted-cost proposed solution validated using several scenarios mimicking real-world deployments two different settings — computer simulations docker testbed. Our empirical evaluations show that total discounted cost incurred by similar state-of-the-art deep algorithms PPO (Proximal Policy Optimization) A2C (Advantage Actor Critic) but requires order magnitude less time train, outputs easily interpretable policy, can be deployed online manner.