Light Reading is asking the question “Where is SDN Going?” today (http://www.lightreading.com/blog.asp?blog_sectionid=388&doc_id=228144&) and I’d like to propose we ask a broader question, which is “Where is the network going?” as we wind down 2012. The reason is that SDN is a network trend being driven by network forces, and you have to get to the heart of the food chain to understand the menu.
We don’t have settlement on the Internet. Given that reality, we don’t have any mechanism to prevent financial arbitrage of Internet transport and connection resources. The landscape of the Internet would be very different if we had settlement, and not “very bad” as most suggest. We’d have less pressure on operators to build out the access network, we’d have less pressure on privacy because we’d not rely as much on ad funding, and we’d have fewer VCs sucking the blood of innovation with mindless startups. But I digress. The point is that absent a mechanism to value bits, bits become non-valuable and that means transport/connection services are loss leaders. That raises two questions; what is the profit source they presume to lead to, and how do we minimize the damage of the loss-leading process.
The cloud is the goal of pretty much everyone these days, and the cloud is best visualized as an IT architecture to deliver the “service” component of network services using the connectivity fabric of the Internet. The two concepts are complementary, in other words. We’re moving toward a cloud-based execution of network services of all types—services above and beyond transport and connection.
There are drivers of change created by the cloud, the most significant of which is the creation of “domains” where connectivity and traffic handling rules are different. In these domains, applications drive the network, and the notion of software-defined networking (SDN) and virtual networking address the issues these domains create. One such domain is the data center, but it’s only a subset of the real domain issue, the “cloud core” where interprocess cooperation composes the experiences we’ll pay for in the future.
On the cost side we also see SDN, driven by another set of issues. Networks, even without the cloud, face a different set of reliability/availability issues today. “Least-cost” routing, for example, has to consider more the cost of disruption than the cost of steady-state operation. With the cloud included we face the reality that the cost of a service and the availability of the experience it brings its users depends more on IT resources than the delivery fabric. Both these forces drive us toward considering the network more as a software resource. They also demand we answer the question “does this allow us to dumb down devices to reduce costs?”
Network Functions Virtualization is a more general response to the cost question. If we have cheap industry-standard servers made cheaper by creating cloud resource pools, why could we not offload anything but the data plane functions of network devices onto these pools? The result would be cheaper handling of the connection-related features of the network. NFV is cloud for connection-related features, in short. It’s also obviously aimed at cutting costs both by reducing the cost to host features and by reducing differentiators that let a given vendor charge more for gear.
SDN can’t move on cost alone, or on cloud alone, and NFV or virtual networking have the same constraint. All of this is aimed at creating a survivable service ecosystem, and that demands a combination of revenue upticks and cost reductions. The network can never be a real loss leader because if it is, operators start their service-layer competition with a deficit they have to make up. That’s a deficit the OTT players would not share, and so they’d have the advantage. No amount of new revenue will stop operators from pressuring vendors to reduce network cost. That’s why Cisco is smart in saying they have to become an IT company. Even in the NETWORK of the future, IT is going to be the value-creating, competitively differentiating, element. It didn’t have to be that way; five years ago operators tried to get vendors to fix their problems. The vendors didn’t respond, and now it’s too late to make the future of the network be about networking. It’s all the cloud now.