Container image management in cloud-edge environments: an image deletion method based on layer affinity

Edge computing, which offloads data processing and analysis tasks to the edge of network, has gradually become a critical technology for handling large-scale data and providing low-latency services. Container technology is a lightweight, portable way of packaging software that is well suited to edge computing environments because it simplifies the deployment and management of applications while reducing the dependence on specific hardware. The synergistic use of container and edge computing has emerged as an ideal approach for achieving efficient and reliable service delivery. When a new container instance is deployed, the corresponding image must be downloaded from a remote image registry. However, edge nodes typically have limited storage resources, which means that storing a large number of images is not feasible. In edge environments with limited network resources, downloading the required images significantly increases container startup time, affecting the quality of service. Current research often focuses on container startup and image distribution, but frequently overlooks the management of storage space for images at edge nodes. We propose Image Cleaning method based on Minimum Layer Affinity (ICMLA), aiming to innovate strategies for managing images storage space in the edge computing environments. To evaluate the efficacy of the proposed method in this paper, a simulated cloud-edge cluster, designed to emulate a real-world operational setting, was established, and a series of comparative experiments were executed to assess performance. Experimental results show that, compared with the typical image replacement algorithms, ICMLA can reduce the total Pods startup time by 13.31% and increase the hit count by 56.67% at most.