The Cloud-Native OVS Datapath Challenge

This repository contains a fully automated, production-grade setup to deploy KubeVirt Virtual Machines attached to an Open vSwitch (OVS) secondary network bridge using Multus CNI. It serves as a foundation for understanding vDPA hardware offload on NVIDIA DPUs.

---

Prerequisites

Before starting, ensure you are running on a Linux machine (Native Ubuntu is highly recommended for KVM hardware virtualization speed) or Windows WSL2 with Docker Desktop.

You only need one thing running:

• Docker (Ensure the Docker daemon is started)

Note: The setup script will automatically detect and download all missing CLI tools (like kind, kubectl, and virtctl) locally to save you time!

---

One-Click Automated Setup

We have engineered an advanced 13-step idempotent pipeline that handles the entire lifecycle automatically:

1. Validation: Checks system resources and prerequisites.
2. Environment: Safely cleans up old deployments and builds a fresh Kubernetes KinD cluster.
3. Network: Installs Open vSwitch on the worker nodes, followed by Multus CNI and the OVS CNI plugin.
4. Virtualization: Deploys the KubeVirt virtualization engine.
5. Workload: Automatically deploys two VMs (test-vm-1 and test-vm-2) and connects them to the secondary OVS datapath.
6. Verification: Uses sshpass to perform an automated guest-level ICMP ping test directly from test-vm-1 to test-vm-2 across the OVS bridge.
7. Evidence Collection: Captures the ping output and extracts the highly detailed OVS flow and MAC learning tables.

How to Run:

Simply open your terminal in this directory and execute:

bash cluster_setup.sh

Sit back and watch the pipeline build your cluster and verify the datapath!

Expected Automated Outputs

Once the script completes, it will automatically generate the required verification files in your folder:

1. ping_results.txt
   • Proves guest-to-guest connectivity with 0% packet loss.
2. verification_flows.json
   • A detailed JSON dump containing the OVS flow rules, bridge configurations, learned MAC addresses, and capture metadata.

(You can view these files directly to verify the success of the datapath!)

Ping_result.txt and verification_flows.json

---

Cleanup

To safely tear down the cluster and wipe the Docker nodes, simply run the built-in self-destruct command:

bash cluster_setup.sh cleanup

---

Architectural Shift to DPUs

To understand why we built this software datapath, and how this exact Kubernetes topology translates to blazing-fast hardware acceleration using an NVIDIA BlueField-3 DPU, vDPA, and OVS-DOCA switchdev mode, please read the included architectural document:

👉 Read the DPU Offload Concept Document