The story that Amazon is looking to get into the white box switch business is probably true at some level. Perhaps they really want to sell them, or perhaps they’re simply looking to design their own hardware for their cloud data centers. In either case, it should indeed make the network vendors who rely on switching (which is most of them) shiver a bit. It also should tell us something about how vendors are approaching their own business.
Back in the ‘80s, I used to teach network salesforces how to sell. One of the mantras was “feature, function, benefit”, meaning that you told a prospect about a feature, you told them what that feature did, and then you told them how it benefitted. The point behind this progression was that people bought stuff because it did something good for them.
A corollary concept to this is the axiom that people will buy the stuff that’s best for them. You can connect the two and say that they’ll by the one whose features create the most benefits, and that would also be true. But suppose the features of every vendor, every product, are the same. What you then buy is the cheapest.
What’s been happening in networking for the past ten years or so is that the pace of creating relevant features has slowed. The new box is pretty much the same as the old, feature-wise. That means that it’s the same as the old, benefit-wise, too. That in turn means that unless it’s cheaper, there’s nothing you can say for it. We’ve fallen into price differentiation because the new features have dried up.
This isn’t the fault of buyers, it’s the fault of sellers. We can’t expect that enterprises or network operators will pay staffs of great thinkers to come up with network features that they hope somebody will build. They’d expect the sellers to do the great thinking, and the sellers have fallen down on the job.
When I started this year, I was working to figure out what the major service and product trends in networking would be. Many of you have read my pieces on logical networking, NaaS, and SD-WAN, all of which came out of this review. The review provided other interesting insights, and one of them was that well over half the people who were actively pursuing network procurements believed that there were network features that would have made their network applications better, but that vendors hadn’t bothered to develop them.
It gets worse. Of that group, almost 80% said that they believed that there were business benefits left on the table because their networks couldn’t address them. Just short of that number believed that the cloud and virtualization demanded a new model of networking. None said they had heard of such a model from their vendors.
The point here is simple. Amazon is hardly a switch player, they’re a mass-market retailer. If they believe there’s an opportunity for them in switching, they believe it’s a mass-market opportunity. No special features, no new applications to teach users—buy a box and plug it in and may the cheapest box win. The only alternative position that would make sense is that switching isn’t a commodity market because there are rich new things that networking could address, and only specialists can help address them. So where’s that story?
The problem here, I think, is connectivity-centricity. I remember when computer-to-computer communications took place at 2400 bits per second. Today, kids would laugh at anything that slow and businesses probably couldn’t even turn on their lights with that kind of connection. But faster connection isn’t necessarily more useful connection. Even twenty years ago, I remember an Intel engineer telling me that in the newest and fastest chips, almost three-quarters of the incremental power was going into making the GUI pretty. That says that the utility of compute power had more to do with presentation than with the quantity and quality of data. Might this also be true of the networks?
I also found in my discussions with users that the term they thought best described the architecture of the future network was network-as-a-service, and as I said in a prior blog, the definition of NaaS they liked best was “a model of network service where connectivity is established where and when it’s needed.”
In our connection-centric mindset, this has been interpreted as meaning “to the sites I need connected”, but you can see that’s not really what users had in mind. There would be little variability in how many sites a company had or where they were located, after all, so the “when” and “where” would be static. Users obviously have something more dynamic in mind.
What? Way back in 1974, there was a specification created called “Basic Reference Model for Open Systems Interconnect”. The “OSI model” had seven layers (often today called “levels”), and almost everything we do today in network equipment stops with Layer 3, which today means IP. Site networking is Layer 3 networking, for example. What’s above that? Is there something that this almost-ancient model got that was lost somewhere along the way?
Layer 4, transport, does exist with TCP, and it handles flow control. Level 5, the session layer, is where the kind of ad hoc relationships between users and resources fits, and remember that seems to be what users think NaaS should provide. Layer 6 deals with information formatting and presentation, and Layer 7 is the application layer where communications control, relationships, and identities reside. The user was presumed to connect up at Layer 7.
Network vendors hunkered down at Layer 3 because it was easier, because site networking was necessary before anything else could be connected, and because the OSI model said that everything from Layer 1 to 3 was “the network” and the rest was in the end-user domain. Not only that, those additional layers were explicitly creating an overlay network.
Today, people in many standards groups are looking at standardizing headers and features that fall within the definition of those higher layers. It’s not that we don’t have people claiming to have “7-layer architectures”. The problem is that they’re inventing the other layers, not doing what the model described.
The solution to competition like Amazon, or to commodity network hardware in general, is to accept that “the network” was supposed to be virtual or logical from the first, from those early glory days of the OSI model. What we think of as “NaaS” is really a modernized formulation of an overlay network model that provides the kind of connectivity users want from the port side, and links to those old commodity site networks on the trunk side.
We could build NaaS with a product family like SD-WAN. We could deploy it in universal CPE (uCPE), in edge computing, in a specialized network device, in a lot of different ways. We could integrate it with traditional Layer 2/3 (switch/router) products, and thus differentiate them. There are plenty of ways that traditional vendors could get into the NaaS space, to elevate networking beyond the commodity level. Sure they might be admitting to their white-box and commoditization problems, but don’t most people see these problems already? Not admitting to morning isn’t going to stop the sun from rising.
I think that something like NaaS, something like the “logical networking” concept I’ve blogged about, is a reasonable, perhaps even optimal, solution to the challenges of networks-beyond-sites, but I’m not promoting that. What I’m promoting is remembering the past. Remember Novell? They were the startup of the age in their time, framing file-sharing and printer sharing and quickly becoming the player in the “network operating system”. In a couple years they were gone, because they couldn’t figure out what came next. Which was virtual resources, virtual machines, containers, and more. Myopia is provably fatal.
Amazon’s white-box aspirations, whatever they are, threaten the current network vendors only because the current network vendors elect to fight head-to-head where there’s no more feature differentiation to be had. The real risk that players like Amazon, Google, Microsoft, and a host of startups pose to the network giants isn’t that they’ll get into the switching business. It’s that they’ll go higher on that old OSI stack and seize the real value.