With the increasing scale and complexity of edge computing tasks, the requirements for storage system performance are becoming higher and higher. However, the current block I/O workload analysis methods are often too rough to accurately capture the diverse and complex I/O operation characteristics in the edge computing environment. This leads to optimization of the storage subsystem based on distorted simulated workloads, which cannot effectively support actual edge applications, resulting in overall system performance degradation. In this paper, we propose an application-grained block I/O workload simulation (AGBS) tailored for the edge computing environments, aiming for fine-grained workload characteristics. The framework structure of AGBS is first described, including trace collection, application isolation, application weight analysis, single application feature analysis and simulation, and multiapplication simulation load fusion. Following that, we discuss and validate multidimensional feature portrayal methods for a single application. Finally, using the AGBS framework, we implement a workload analysis simulator. Experiments show that the simulated I/O workloads based on the AGBS framework are stable within 40% error of the actual workload in terms of workload stress, autocorrelation, and request length; in the best outcome, the combined error is less than 5%.

The deployment of increasingly diverse services on edge devices is becoming increasingly prevalent. Efficiently deploying functionally heterogeneous services to resource heterogeneous edge nodes while achieving superior user experience is a challenge that every edge system must address. In this paper, we propose a Container-Node Graph (CN-Graph) based container optimal matching deployment algorithm, Edge Kuhn-Munkres Algorithm (EKM) Based on Container-Node Graph, designed for heterogeneous environment to optimize systemperformance. Initially, containers are categorized by functional labels, followed by construction of a CN-Graph model based on the relationship between containers and nodes. Finally, the container deployment problem is transformed into a weighted bipartite graph optimal matching problem. In comparison with the mainstream container deployment algorithms, Swarm, Kubernetes, and the recently emerged ECSched-dp algorithm, the EKM algorithm demonstrates the ability to effectively enhance the average runtime performance of containers to 3.74 times, 4.10 times, and 2.39 times, respectively.