MEC Resource Orchestration for Heterogeneous Networks and Services Using Reinforcement Learning

Rapidly changing wireless network paradigms and technologies that can serve diverse applications and services have become the focal point. Multi-access Edge Computing (MEC) offers promising solutions to mitigate the complications created by such dense networks within the Management and Orchestration (MANO) frameworks. The MEC is expected to handle demands from different services that have heterogeneous performance constraints. Reinforcement Learning (RL) technique is used to solve the resource orchestration and allocation tasks at the MEC in this paper. Soft Actor-Critic (SAC) is used as the RL agent to learn the local network resource orchestration model and is further used for resource allocation also. Network resources such as CPU and memory are considered within the model for meeting the required latency of heterogeneous network services. The simulation results prove that the SAC based resource orchestrator gives a better performance in comparison to the random service orchestrator. The prospects of using such technique for multitude of use cases in Future Wireless Networks (FWN) appears to be indisputable.