Integrating Tungsten Fabric, OVS and SR-IOV compute nodes through an OpenStack ML2 mechanism

The challenge

OpenStack users already ran workloads on compute nodes based on OVS and SR-IOV. To bring them to Tungsten Fabric, those nodes had to work alongside vRouter-based computes under one management point, and migrate to vRouter live. This R&D project had to produce a Proof of Concept in time for a demo at the 2019 Open Infrastructure Summit in Shanghai.

What was at risk:

• A fixed conference deadline for the PoC, and OpenStack users whose existing OVS and SR-IOV compute nodes had to fit with Tungsten Fabric

What we did

• Provided a team of software and network engineers with experience in developing Tungsten Fabric, each with skills across hardware, networking, software and DevOps.
• Prepared a Proof of Concept for a demo at the 2019 Open Infrastructure Summit in Shanghai, delivered in short iterations so the client could adjust the scope and requirements.
• Built the solution on the ML2 MechanismDriver framework from OpenStack.
• Delivered CI changes and a complete set of automated tests.
• Supported the client with a live PoC demo at the Open Infrastructure Summit.
• Released the solution to the open-source community.

How it works

The solution builds on the ML2 MechanismDriver framework from OpenStack, which lets numerous ML2-based SDN drivers work simultaneously. Three things follow from that.

1. One central point Workloads are managed and orchestrated from one central point, and security policies are reused across them.
2. Mixed workloads OVS-, SR-IOV- and vRouter-based workloads run simultaneously, with OVS and SR-IOV workloads running while Tungsten Fabric manages the underlying fabric.
3. Live migration OVS-based computes can be live migrated to vRouter-based computes.