I just met up with DELL guys for Big Switch SDN. They claim there is no routing running on leaf switches, the BCF is purely OpenFlow.
Want to hear your idea how they can handle network partitions?
They can’t (at least not to the same extent the SDN solutions that don't have centralized control plane like Cisco ACI can). They have a “headless mode” where the network keeps forwarding stuff even when the switches lose connectivity to all controllers, but the moment you have a topology change, which can be as simple as a VM (and thus MAC/IP) move, you’re toast.
In theory, you could cope with core link failures in headless mode if (A) the switches run some link quality monitoring mechanism (example: BFD) and (B) the controller pre-programs backup paths into the switches. There’s no way to cope with external changes without access to a proper control plane.
I think the leaf switch do not have any control-plane.
Somewhat correct. They have ARP and LACP offloads, and they’re probably running BFD on the core links, so at least these things continue to work when the controller is not reachable.
You’ll find a few more details in the SDN Use Cases webinar (I spent ~15 minutes explaining BCF), and if you’re interested in designing more down-to-earth data center infrastructure, check out my online course.