PiPD-CIRCLE: Pipeline for Predictive Distributed Container Image Registry with Cloud Backup and Lazy-Pulling in Edge Environment
Abstract
The significance of edge computing is soaring in the rapidly evolving era of 5G/6G networks. Enhanced edge network capabilities pave the route for applications such as autonomous vehicles, telemedicine and remote patient monitoring. Given the precarious nature of edge networks, deploying software via containers has become a practical choice. Most applications require image-pulling for container startup, for which the required docker images are typically stored in a central cloud registry. Restricted bandwidth between the edge and cloud networks causes latency during the image-pulling procedure, creating a bottleneck and leading to prolonged container startup times. On the other hand, image push in edge environments poses security threats and stale states, leading to attacks and malicious content. Therefore, in this paper, we propose an end-to-end pipeline called PiPD-CIRCLE for image push and pull supported by a distributed container image registry based on peer-to-peer connections of edge nodes and a robust cloud backup to reduce container startup time. Our cloud backup enables high availability and ensures the integrity of content entering edge nodes via vulnerability scans and sign verification. Additionally, our design incorporates the concept of lazy image pulling and an AI module to predict image placement on edge nodes for handling cold starts efficiently. Experimental evaluation shows that we have achieved over 70% reduction in container startup time compared to pulling from the central Docker Hub registry while also enabling the tagging of vulnerable images at the edge entry point to mitigate security risks.