Multi-Agent Reinforcement Learning for Workload Distribution in FaaS-Edge Computing Systems

Edge Computing has emerged as a response to the growing adoption of the Internet of Things, reducing latency and enabling real-time data processing by shifting computation from centralized cloud servers to the network edge. Adopting the Function-as-a-Service model at edge nodes would further increase flexibility and cost-efficiency, but comes with several challenges related to resource constraints and unpredictable traffic patterns. In addition, efficient task offloading approaches are critical in distributed edge environments. Reinforcement Learning (RL) can be beneficial in this context, outperforming traditional heuristic methods due to its ability to dynamically optimize workload distribution. In this preliminary study, we investigate the potential of multi-agent RL for workload management in federated FaaS-Edge environments. By comparing the widely-used Proximal Policy Optimization (PPO) and Soft Actor-Critic RL methods with two heuristic baselines, we highlight how PPO holds promise in increasing the processed load by exploiting requests forwarding to neighboring nodes.