Optimal Resource Management for Multi-access Edge Computing without using Cross-layer Communication

We consider a Multi-access Edge Computing (MEC) system with a cloud server, a base station (BS) and an MEC server attached to it. The resource constrained MEC server can be dynamically configured to serve different classes of services. The users send all the service requests to the BS which in turn keeps a subset of the requests to be served at the MEC and forwards others to the cloud server. The service requests that are processed at the MEC server incur queuing and processing delays whereas those that are sent to the cloud only incur fixed processing delays. Throughput and delay optimality warrant uplink packet scheduling at the users, MEC server configuration and service scheduling at the BS, and service forwarding to the cloud accounting for the system state. Traditional solutions to this resource management problem, e.g., those based on back-pressure, entail cross-layer message exchange. We develop two virtual queue-based drift-plus-penalty algorithms that do not require cross-layer communication, are throughput optimal, and achieve the optimal delay arbitrarily closely. The algorithms offer a tradeoff between the queuing and processing delays at the MEC server and the service processing delay at the cloud. We illustrate the performance of the algorithms via simulations.