New Cache Hierarchy for Container Images and OCI Artifacts in Kubernetes Clusters Using Containerd

Learn how to implement a new cache hierarchy system for container images and OCI artifacts in Kubernetes clusters using containerd in this 31-minute conference talk from KubeCon + CloudNativeCon. Discover how Toru Komatsu and Hidehito Yabuuchi from Preferred Networks, Inc. developed an innovative solution to address one of the key bottlenecks in Kubernetes pod startup: the time required to pull container images and OCI artifacts from registries. Explore the three main features of their cache system including a new cache hierarchy that enables cluster-wide image sharing and optimization beyond traditional cluster-local caching, a "Ninja" approach that transparently enhances performance without requiring user modifications, and preheating capabilities that support pushing images to warm the cache for faster subsequent pulls. Examine real-world production results showing approximately 95% cache hit rates that significantly reduce pod startup times and minimize network communication with registries. Gain practical insights into leveraging cache mechanisms and Container Runtime Interface (CRI) to optimize image and OCI artifact pulls, ultimately improving overall cluster efficiency and reducing operational costs associated with frequent large container image fetches.