Abstract
Edge computing reduces the response time of real-time services with handling dynamic traffics reliably. Edge servers with limited resources often utilize container technology, which provides a lightweight execution environment. When deploying containers on edge servers, a container image is required and it is predominantly downloaded from a remote registry. Therefore, these operations are heavily influenced by network overhead between the container deployment system and the remote registry. Through motivation experiments specifically designed to identify these network overheads, we demonstrate that container pooling time increases in proportion to the physical distance between two hosts and varies flexibly based on runtime conditions. In this paper, we define a system model to place multiple registries for high-speed container deployment and propose a technique for clustering edge servers and selecting leaders within each cluster based on the affinity between edge servers and network overhead. Additionally, we propose a technique for selectively deploying registries, taking into account the idle resources of each edge server. A simulation experiment is conducted to verify the performance of the proposed technique and shows that performance improvement can be achieved regardless of the number of edge servers and the k value.