The “Silk” Road: Building a CNI Plugin from Scratch - Usha Ramachandran & Angela Chin
summary
Built a bespoke CNI called Silk that allocates a unique IP per container using a VXLAN overlay and a per-host lease model: a silk controller hands out per-cell subnet leases, a daemon renews them and the CNI binary assigns container IPs. Design rules: don’t acquire a new lease if containers exist, fail networking if no lease, and preserve leases across partitions; teardown clears stale state. Bandwidth limiting is extracted as a bandwidth CNI plugin and chaining is supported; Kubernetes use is experimental.
AI-written summary · about
chapters (15)
- 0:00Intro
- 1:36What is CNI?
- 6:20Cloud Foundry Application Runtime
- 7:12Cloud Foundry Workflow
- 9:23Cloud Foundry Legacy Networking.
- 10:26Networking Requirements
- 11:59Connectivity - Requirements
- 12:23Connectivity - Solution
- 15:28Extensibility
- 16:22Building Silk
- 17:27Reliability & Stability
- 22:52Network Setup
- 24:59CNI Chaining
- 26:57The Future of Silk
- 27:39Key Takeaways
related talks
37mIntroduction to CNI, the Container Network Interface Project
33mKubernetes Networking: How to Write a CNI Plugin From Scratch
34mHeterogeneous multi-cluster full mesh communication practice
40mModular Overlay Networking Solutions with the Container Network Interface
31mEverything You Ever Wanted to Know About CNIs (Container Network Interface) and ...
38mVPCs in Kubernetes
32mDemystifying CNI Writing a CNI from Scratch
46mMulti - Networking Kubernetes Containers with CNI