RDMA Enabled Kubernetes for High Performance Computing
Talkat KubeCon + CloudNativeCon North America 2019 · on Thursday, 2019-11-21· at 16:25· Room 33ABC - San Diego Convention Center, Upper Level·
summary
Describes a Kubernetes setup that wires RDMA-capable NICs into pods using Multus CNI plus a custom SR-IOV device plugin and SR-IOV CNI, coordinates with CPU Manager and Topology Manager, and runs 200 iterations of RDMA benchmarks on Mellanox hardware (40 Gb/s link). Claims RDMA bandwidth ~4 GB/s in containers, within 1% of bare-metal, and about 40x higher than default Flannel (<1 Gb/s). Provides YAML snippets, pod annotations, BIOS/hugepage steps, and a working demo—RDMA in Kubernetes works and matches native performance.
AI-written summary · about
chapters (16)
- 0:00Intro
- 0:50What is High Performance Computing and why is it important?
- 2:57How can Kubernetes help HPC?
- 6:47Networking in Cloud Computing
- 9:39Networking in HPC - ROMA
- 11:14Implementation
- 12:14Kubernetes Networking
- 13:11Multus
- 15:15Enabling SR-IOV and RDMA
- 17:22CPU Manager and Topology Manager
- 19:54RDMA Workload Pod Spec
- 21:12Lab setup
- 21:51Kubernetes setup
- 22:32Config Map specifies the details of SRIOV capable NIC
- 28:27Conclusions
- 29:11Future Work
related talks
26mAdvanced Network Capability for NFV on Kubernetes
25mHigh Performance Networking for Distributed DL Training in Production K8s
27mMake Kubernetes Networking Ready for World Class AI and HPC ...
31mBreaking the Single-Network Barrier: A Cross-Domain Look at…
34mUnlocking Performance: Topology-Aware CPU Scheduling W...
35mMigrating the Next Generation Mobile Core Towards 5G with Kubernetes
37mHigh Performance Networking with Kubevirt
51mAccelerate containers and VMs networking with vDPA