The notion of “revolution” is always exciting, sometimes useful, occasionally destructive. The notion of two or three of them at once tips the scales into the latter category in my view. We have been looking at “the cloud revolution” for several years, we’ve just started “the SDN revolution” and now we’re facing “the NFV revolution”. I propose that the latter two are driven by the first, the cloud, and that these two are destined to become one. So while we need to understand each of our “revolutions” the most important thing is understanding how they’re joining up. Because, friends, the future network is ONE NETWORK and not three.
The cloud defines a new relationship between applications and resources, and thus defines a new mission for networks. All revolutionary changes demand a massive influx of benefits to drive them, and the cloud is the embodiment of these new benefits. No, it’s not the crap about IaaS displacing internal IT, but the way the cloud defines the fulfillment of point-of-activity intelligence—a market that holds over two trillion dollars in incremental service revenues.
An immediate result of cloud consideration is SDN. I’ve blogged before that we have three “accepted” models of SDN. One is the OpenFlow “purist” model that actually replaces adaptive discovery to build forwarding tables with explicit central control. One is the “virtual network model” that abstracts the connection network above Level 3 and thus disconnects it from real devices (vSwitch). One is the “distributed” model that controls the network not with OpenFlow but with protocols/standards that evolve the current Ethernet and IP. If you attempt to extract a sense of mission from this combination, you find the only really common link is “the cloud”. All of the models of SDN presume a tighter linkage between network and application processes to facilitate a new union of IT and networking. There’s a cloud above everyone’s view of SDN.
The emerging notion of Network Functions Virtualization (NFV) our most recent “revolution”. The basic notion of NFV is simple; separate the logic associated with a higher-level service from the data-plane connection that delivers it. Network functions are virtualized, meaning made platform-independent, and then hosted in a resource pool. That pool would, to most of us, be “a cloud”. The purpose of NFV is to decouple network features from monolithic devices and in doing so make the network more responsive to changes in mission, which would of course represent changes in market needs. So there’s a cloud above NFV too.
If SDN and NFV are both cloud-driven, are they convergent? If we presume that the “software definition” that’s driving the “network” in SDN is a set of application functions that are today at least implicitly supported in the network devices, then NFV is a superset of SDN. In my view, SDN is a subset of NFV that’s focused on the specific “service” of topology management and route selection. That there is an overlap at least is acknowledged by the NFV white paper. So let’s look out past the inevitable (to me) convergence of SDN and NFV and focus on the clearly more complicated goal of NFV. It postulates the offloading of “service logic” from network appliances to be hosted on servers.
To me, this goal begs two questions. First, what will define the architecture of the platform-as-a-service that hosts this functionality? You can’t write software in a vacuum; we can’t use different principles to unload every possible network feature from devices to host it in the cloud. Second, what’s left in the network when you remove the “virtual functionality”? You can never dispense with the network as a collection of devices, so in the NFV world of the future, what exactly are those devices? If the goal of NFV is to simplify network devices and make them more flexible and responsive, you’d have to assume that the network device is changed as a result of removing of the virtual functions.
This whole SDN/NFV thing is critical for “networking” because whatever you believe about their relationship the two show we are transitioning from a world of “IT and network” to a world of “IT/network” that we’ve called “the cloud”. Where the IT/network balance will fall, both in terms of technology contribution and in terms of future capex, will determine the fate of network vendors, network operators, and of course network professionals.
Out in the great beyond, where will cloud networking, SDN, and NFV combine? In a new service layer that’s cloud-like and a new device that’s not switch or router but something more? I’ll be rolling out the general requirements of that device in this blog too, and I’ll be interested in what you think.