LEASE: Leveraging Energy-Awareness in Serverless Edge for Latency-Sensitive IoT Services

Resource scheduling catering to real-time IoT services in a serverless-enabled edge network is particularly challenging owing to the workload variability, strict constraints on tolerable latency, and unpredictability in the energy sources powering the edge devices. This paper proposes a framework LEASE that dynamically schedules resources in serverless functions catering to different microservices and adhering to their deadline constraint. To assist the scheduler in making effective scheduling decisions, we introduce a priority-based approach that offloads functions from over-provisioned edge nodes to under-provisioned peer nodes, considering the expended energy in the process without compromising the completion time of microservices. For real-world implementations, we consider a testbed comprising a Raspberry Pi cluster serving as edge nodes, equipped with container orchestrator tools such as Kubernetes and powered by OpenFaaS, an open-source serverless platform. Experimental results demonstrate that compared to the bench-marking algorithm, LEASE achieves a 23.34% reduction in the overall completion time, with 97.64% of microservices meeting their deadline. LEASE also attains a 30.10% reduction in failure rates.