It’s always fascinating to listen to network operators and even large enterprises talk about their infrastructure projects, and I’m just starting to analyze the first round of our fall strategy survey, so I’m getting that chance on a large scale. It’s too early to say how everything is going to come out, but one thing that strikes me at this point is the discrepancy between how real network-builders see their networks and how vendors see them.
In the service provider space, the classic vision of networks is the hierarchy — access to aggregation/metro to core. The goal is supposed to be the creation of uniform connectivity and good bandwidth economy of scale, and the practice dates from the earliest days of packet networking. It’s also nearly always at odds with what’s going on in the real world. Today, network hierarchy is being replaced with delivery. It would be safe to say that were video content the only traffic source (or even the overwhelming majority of traffic), the Internet would look more like a content deliery network (CDN) than a hierarchical network. All traffic would be user-to-cache and the only “core” traffic would be to populate caches.
In the enterprise space, people are starting to realize that the applications that consume a mass of incremental bandwidth aren’t universally distributed through the business. Enterprises tell us that 73% of all their data center traffic and 81% of their collaboration traffic moves less than 10 miles. The data center network is growing very fast as inter-process and storage traffic multiply, and LAN traffic in major headquarters facilities is growing nearly as fast, but branch traffic is growing at a much slower rate.
You need only reflect on what happens in a local sales office or branch bank to understand why. Most remote-office traffic is transactional and thus doesn’t expand with anything but business activity. Who does a bank teller collaborate with if not the teller in the next station (who is hardly a communications destination)? Who does a real estate office manager or manufacturers’ rep communicate with that would necessitate telepresence?
It’s never a good idea to be disconnected from the market reality. We can’t build optimum networks for anyone without understanding what their networks are really doing, and it’s not a problem confined to the network space. Enterprises tell us that vendors are proposing data mining benefits across an employee population whose job activity doesn’t involve any of the data they’re proposing to mine. Providers tell us that vendors are suggesting operations cost savings that exceed their total operations budgets because they don’t understand how much is really spent on operations, or even what the word means to the network operators themselves. Then there’s the question of what new services will do to the space overall.
Appliance players like Apple, Google (via Android), Microsoft, and the host of tablet and phone vendors are creating a more responsive consumer playground to cater to instantaneous trends, or to launch those trends where possible. With a low-inertia alternative developing, we could end up with traditional networking and even IT simply getting out-danced. You can see already that metro networking is becoming Ethernet networking. Alcatel-Lucent’s recent super-switch shows that vendors realize that transport/connection is getting pushed to lower OSI layers, but do they realize that differentiation/innovation is harder to deliver there? The more we want to dazzle the consumer, the more we need to create new in-network value, the more we need to examine a new conception of what a network is and what it contains.