Network Bandwidth-Aware Kubernetes Cluster
summary
Prototype extends Kubernetes with a scheduler extender plus a node agent that labels nodes with max capacity and calls the Cluster API or vSphere provider to reserve bandwidth; uses tc/OVS for per-pod ingress/egress shaping and reports requests like a 2Gbps reservation. A worked demo schedules around a 4‑node example (three ~1Gbps nodes, one low‑capacity node) by filtering out insufficient nodes and scoring on remaining bandwidth. Implementation functions end‑to‑end but is conservative: reservations are compressible and cannot fully guarantee cross‑node contention.
AI-written summary · about
chapters (9)
- 0:00Start
- 3:42Kubernetes Standard Scheduler
- 6:40Network Features in Kubernetes (2/3)
- 13:23BareMetal Deployment
- 18:36Reserve Bandwidth on Workers
- 21:45Mark Pods with Resource Request
- 24:31Algorithm for Extended Rules
- 26:16Extended Scheduler
- 27:35Demonstration
related talks
36mNetwork Bandwidth-Aware Kubernetes Cluster
36mNetwork-aware Scheduling in Kubernetes
30mSched.net: A Network-Aware Kubernetes Scheduler [I]
34mNetwork QoS Support for Kubernetes Applications
32mCoordinate Workloads Colocation: QoS-Oriented Scheduling Enhancement on K8s
30mEnhancing the Kubernetes Scheduler for Diverse Workloads in Large Clusters
27mCloudMirror: Application-Aware Bandwidth Reservations in the Cloud
38mDynamic Pod Resource Boundary Adjustment in Web Scale Clusters