Introducing Centaurus: A Cloud Native Infrastructure Project for Large Scale Distributed Cloud
summary
Claims a Kubernetes-derived compute plane that scales to 50k nodes per cluster today with a goal of 300k nodes, adds a unified API and runtime for VMs and containers (pods can be VMs), and implements multi‑tenancy via isolated “spaces” that act like virtual clusters. Networking uses a high‑performance XDP/eBPF dataplane (Mizar) with VPC/subnet primitives and per‑node XDP programs; edge stack supports hierarchical, multi‑flavor clusters for low latency and autonomous operation. Performance tests report 1.5M containers and 6x QPS improvement versus stock Kubernetes.
AI-written summary · about
chapters (18)
- 0:00Start
- 4:16Agenda
- 4:56What Is Centaurus
- 6:12Manage Distributed Cloud at a Large Scale
- 12:23Centaurus Edge
- 12:39Key Features
- 15:04Unified Elastic Framework
- 15:37Key Highlights
- 18:17Major Changes That We Made to the Kubernetes
- 19:57Kubernetes Architecture
- 21:49Resource Partition
- 30:56Network
- 31:34Network Objects and Network Controllers
- 34:50Runtime Infrastructure
- 39:14Scalability
- 43:26Kubernetes Cluster
- 52:49Goals and Visions of the Edge Computing
- 58:45Key Partners
related talks
39mThe Multi-Tenant Supercluster Approach
05mkcp: Scaling The Kubernetes Control Plane For The Multi-Clust...
31mAn Adventure in Multi Cloud with Kubernetes and Distributed Data
34mScaling Kubernetes up to 1,000 Control Planes - Dario Tranchitella
16mHow We Built a Cloud On K3s: The Learnings Of Growing Up Fast
50mScale Kubernetes to Manage 60K Nodes Through Architectural Extension
26mKubernetes - The Universal Control Plane
37mKubeInception: running up to a thousand clusters in the real Cloud-Native way