EdgeOptimizer: A programmable containerized scheduler of time-critical tasks in Kubernetes-based edge-cloud clusters

Edge computing has garnered significant attention in recent years, leading to the evolution of more delay-sensitive applications towards a three-tier architecture with cloud–edge collaboration. Concurrently, technologies associated with containerization have been maturing. Notably, K8s(Kubernetes) emerges as a prominent solution for the management of extensive, dynamically evolving, and intricate container clusters. However, optimizing performance in a K8s-based architecture requires careful consideration, as intelligent algorithms cannot be easily rehearsed or retracted, and a poorly functioning algorithm can result in significant damage. To enhance the success rate of time-critical task execution in real production environments, it is crucial to provide an algorithm optimizer for service orchestration and request dispatching, especially when dealing with different services. Traditionally, building a K8s-based experimental system during the early stages of experiments has been time-consuming and involved significant programming efforts. In this paper, we introduce EdgeOptimizer, a decoupled and modularized optimizer for scheduling in multiple clusters. EdgeOptimizer also serves as an online testbed for algorithm verification in K8s-based systems, offering detailed configuration options to facilitate cluster management. By collecting system information and employing an interface-oriented system architecture, EdgeOptimizer enables users to quickly develop, deploy, and switch between various algorithms, significantly reducing the upfront costs associated with setting up an experimental environment. We evaluated the performance of EdgeOptimizer based on overall overload and demonstrated its scalability and effectiveness in verifying the efficacy of service orchestration and request scheduling algorithms for time-critical tasks. Our findings illustrate the value of EdgeOptimizer in improving the overall success rate of executing time-critical tasks, thus highlighting its potential in real-world scenarios.