Case Study: Preferred Networks unifies Kubernetes networking and boosts AI cluster performance with Cilium from CNCF

Preferred Networks standardizes 5 Kubernetes clusters with CNCF Cilium

Preferred Networks, a leading Japanese AI company, faced a networking challenge in its hybrid Kubernetes environment. While most of its clusters ran on Cilium, one on-premises cluster used Calico, which created a BGP conflict with the host's existing routing process. This forced the team to use inefficient tunneling, which hurt performance and made failures hard to diagnose. To unify its infrastructure, Preferred Networks partnered with the Cloud Native Computing Foundation (CNCF) project, Cilium, for a solution.

The Cloud Native Computing Foundation's Cilium provided a direct routing solution that worked alongside the host's BGP process without conflict, eliminating the need for tunneling. This improved network performance and visibility, resolved a persistent DNS issue, and provided advanced network policy capabilities. Standardizing on Cilium across all five clusters simplified operations and improved stability, allowing the infrastructure team to focus on other priorities while supporting over 1,500 GPUs and 200 custom accelerators.

Preferred Networks