Reinforcement Learning based Cloud and Edge Resource Allocation for Real-Time Telemedicine

Abstract

Future healthcare services will extensively exploit wireless telehealth solutions in various healthcare use cases from preventive home monitoring to highly demanding real-time scenarios, such as monitoring an emergency patient's vital functions in an ambulance or ICU unit. Reliable real-time communications and computing are needed to enable these highly critical health services. However, the majority of current telehealth use cases are cloud - based, which poses a challenge to provide sufficient Quality of Service (QoS). The traditional centralized cloud infrastructure cannot meet the latency and reliability requirements due to long and unreliable communication routes. Therefore, the most advanced cloud solutions integrate edge computing as an integral part of the computational architecture to bring a part of the computational infrastructure to the proximity of the data sources and end-nodes, thus constituting an edge-cloud continuum. This continuum is capable of serving applications with real-time requirements. However, since edge computing capacity is a limited resource, solutions are needed for deciding which tasks should be run on edge and which at the data center. In this paper, we propose a machine learning-based solution to prioritize ultra-low-latency tasks for running on the edge to meet their strict delay requirements while leaving other tasks to be executed at remote servers. Our proposed solution in comparison to the baseline has a significantly lower dropping rate and outperforms fixed - interval scheduling solutions in terms of resource efficiency.