Microservice Provisioning in the Edge-to-Cloud Continuum using Multi-Agent Multi-Armed Bandits

Industrial facilities are increasingly moving towards integrating sensors and actuators to enable real-time decision-making. The realization of this vision calls for the use of cutting-edge approaches at both the infrastructure and application levels. In the former, the Edge-to-Cloud Continuum has emerged with the idea of distributing computing resources across multiple tiers, opportunistically leveraging the proximity of edge servers to deliver low latency and the scalability of cloud data centers to handle large volumes of data. In the latter, microservices provide a flexible, lightweight way to deploy and manage complex applications. Despite the growing research interest in coordinating these two trends, existing solutions typically do not address certain key aspects. First, they overlook that individual microservices can be shared by multiple machines located at various industrial sites, each with distinct latency needs. Second, they fall short in jointly reducing microservice latency and infrastructure power consumption. To address these gaps, we introduce a novel Multi-Armed Bandit (MAB) algorithm that optimizes microservice provisioning in the Edge-to-Cloud Continuum that considers these two aspects. Simulations based on real-world specifications show that our approach reduces latency by 72% and power consumption by 19% compared to a Pareto-based MAB baseline.