Case Study: Kuaishou achieves faster, more stable image distribution with Cloud Native Computing Foundation's Dragonfly and Nydus

Stable and efficient image distribution for 1 billion monthly users with Dragonfly

Kuaishou, a Chinese short video platform serving over one billion monthly users, faced significant challenges with the stability and efficiency of image distribution on its container platform. This was critically important for supporting rapid business growth and large-scale service deployments. They sought a solution to handle the immense bandwidth pressure and ensure seamless service updates from the Cloud Native Computing Foundation's open-source project, Dragonfly.

By implementing Dragonfly and its sub-project Nydus, Kuaishou created a high-performance, P2P-based image distribution system. The solution from the Cloud Native Computing Foundation dramatically reduced bandwidth pressure on their Harbor registry by over 70% on average and 80% at peak. This resulted in a 90% decrease in image pull costs, a 50% saving on pod instance pipeline costs, and enabled tens of thousands of instances to pull images simultaneously without increasing time costs or disk load.

Kuaishou

Hui Zhou

Senior Software Engineer