Deep Reinforcement Learning Approach for Enhancing Profitability in Mobile Edge Computing

With rapid advances in User Equipment (UE) and 5G networks, managing diverse user service needs in densely populated areas presents resource challenges. Mobile Edge Computing (MEC) and network function virtualization are pivotal for Service Providers (SP) aiming to maximize profitability, meet real-time UE resource demands, and adhere to Service Level Agreements (SLA) with minimal resource costs. Our article proposes a request admission-based resource allocation mechanism in MEC networks to address these challenges. Ensuring controlled and secure end-to-end network slicing within an unknown arrival rate time frame, each UE request specifies resources with minimal cost to mitigate SP cost, maximize profit, and adhere to SLA. Admissions rely on secure historical traces during each time slot, considering available resources. Leveraging abandoned instances, remaining capacity, and new instance creation at cloudlets due to resource scarcity, we introduce a deep reinforcement learning algorithm to address heterogeneous UE demands and prevent resource wastage. Simulation experiments confirm the model’s effectiveness in accepting network slice requests based on resource availability, maintaining SLA, and efficiently maximizing SP profit with minimal cost.