Def-DReL: Towards a sustainable serverless functions deployment strategy for fog-cloud environments using deep reinforcement learning

Abstract

Modern cloud applications are composed of tens of thousands of environment-agnostic serverless functions that can be deployed in either a fog or cloud environment. The key to sustaining fog computing is to offload the maximum amounts of computation to the cloud, and accommodate as many users as possible without compromising quality of service (QoS). However, recent research mainly focuses on assigning maximum resources to serverless applications from the fog node and not taking full advantage of the cloud environment, leading to a lack of sustainability in fog computing. As a way to fill this research gap, we explored what percentage of a user’s request should be handled by fog and cloud. As a result, we proposed Def-DReL, a Systematic Deployment of Serverless Functions in Fog and Cloud environments using Deep Reinforcement Learning, by taking into account several real-life parameters, including distance from a nearby fog node and latency, priority of the user, priority of serverless applications, and resource usage. Def-DReL’s performance is further compared with that of recent related algorithms. Simulation and comparison results clearly demonstrate a lesser number of serverless functions from each user (with approximately 10% improvement) being deployed in the fog node, resulting in accommodating limited fog resources to more number of users. The other simulation results show its superiority over other algorithms as well as its applicability to real-life scenarios.