Using Kubernetes with Data Processing Units to Offload Infrastructure
Talkat KubeCon + CloudNativeCon North America 2021 · on Friday, 2021-10-15· at 15:25· 411 Theater + Online·
summary
Describes a working prototype that puts infrastructure services onto PCIe DPUs (32–64 GB, ARM processors) and runs a vendor-neutral CNI offload plus tracing (Jaeger) and Envoy-like service mesh on the DPU. Uses a P4 engine, DPDK and SPDK to move packet, storage and crypto work off x86, exposing APIs via an infrastructure abstraction layer. Result is lower host CPU usage, richer telemetry the scheduler uses for placement, and demonstrable isolation and protocol acceleration.
AI-written summary · about
chapters (8)
- 0:00Intro
- 0:55Foundational Trends in IT
- 6:12What capabilities do we need
- 12:23How it works - DPU
- 15:04How it works - Options
- 17:10Functional Software Architecture
- 20:20Next Steps
- 21:36Summary - 11
related talks
38mKubernetes Networking Infrastructure Offload
32mSoftware Defined Hardware Accelerated Data-Centric Computing...
36mRunning High-performance User-space Packet Processing Apps in Kubernetes
51mAccelerate containers and VMs networking with vDPA
36mBringing vDPA to Life in Kubernetes
05mTowards a Sustainable World Realized by Kubernetes and Composable...
26mMake Kubernetes Greater for NFV@Arm
24mPropagating Programming Paradigms: Lifting High Performance Compute Into K8s