Cisco’s comments on its new “Application-Centric Infrastructure” vision is yet another proof point for my argument that Cisco has successfully played the weaknesses of SDN players against them. In military tactics, seizing the high ground is a maxim. So it is in marketing, and Cisco has done that with finesse.
You can harp on about the value of SDN in terms of efficiency or better hardware/software innovation or whatever, but to actually defend SDN’s value you have to get back to the place where software starts defining networking. That means you have to face those northbound APIs and the applications that control connectivity and manage traffic. In the SDN space, startups have tended to either jump into the overlay virtual networking space or the SDN controller space. The former disconnects the application from the network through an intermediary virtual abstraction and the latter is too low on the totem pole to understand services and support software control. Cisco, I suspect, knew all along that the OpenFlow/SDN community would take too little a bite to be a threat, and they went for the APIs instead.
What Cisco now seems to be doing is preparing a spot for its own SDN-spin-in story, Insieme Networks. That’s actually going to be the tricky part for Cisco, because up to now their SDN approach was the “quacks-like-an-SDN” model; if something exhibits expected SDN properties at the API level, then it’s an SDN. That works as long as you don’t look inside the SDN black box Cisco has built, but Insieme forces Cisco to open that box a bit, to define internal structure to their picture. That will then let competitors take shots not at the philosophy of the Cisco approach but at the technology. So Cisco has to defend.
If you look at Cisco commentary on Application-Centric Infrastructure, you see that there’s a lot of integration and operationalization inside it. It’s tempting to see Insieme as something that would address that, particularly since Cisco is aiming it at the data center. That would make Insieme very Contrail-like or Nuage-like, perhaps, a means of linking virtual and real networks. But virtualization and abstraction work against integration and operationalization, and Cisco will have to address how those two are resolved or face the risk of a competitor who has tangibly better answers.
Network operators have already told me that they’re more worried about how cloud applications (including NFV) are operationalized than about how they’re optimized and deployed. The challenges of operations in a virtual, integrated, world are formidable because there are no real devices that present MIBs, and there are elements of application/service performance that don’t belong to any component that you’ve deployed, but to the connections between components. In the world of the supercloud, the future, you have to be able to derive operations rather than apply them, because in the virtual world there’s nothing real to manage. As we evolve more to virtual elements, we’ll have to face the transition from real device management.
To what? One thing that’s clear is that you’ll need to rely more on automated processes and less on human practices in the cloud of the future. You’ll also have to take a more service-centric or user-centric view of resources and behavior rather than a device-centric view, because you don’t have real devices any more but you’ll always have real users (or you starve, and your problems become irrelevant in a market sense). As an industry, though, we have never really come to terms with a service-user-centric vision of network or IT management; everything ends up coming down to operations centers drilling down through layers of devices.
Whether Insieme contributes anything directly to this process isn’t the relevant point; the challenge is that when you solve problems in a new way you have to operationalize for that new solution. Cisco is dragged into the operations side of the cloud whether they like it or not, and of course so are their competitors. Every SDN strategy should be judged in part based on its operations context, but we’ve been unable to compare SDN operationalization because the competitive focus (and thus the product design and articulation) has been on network features. And Cisco, by abstracting the network, has been able to stay out of the fray completely. Now, with Insieme coming along, it has to dive in, and that makes the full picture from APIs to technology fair game for competitive byplay. Including operations. Especially operations, in fact, because if you can’t carefully operationalize our new and agile virtual world you’ve only invented a new way of getting into deep cost trouble down the line. The complexity of a virtual system is inherently higher because of its additional flexibility and the multiplication in the number of elements.
So watch Cisco’s Insieme stuff for operational clues, and start looking at SDN stories for their operationalization story. What you can’t build and deploy and sustain, you can’t bill for and profit from.