The optical networking conference this week is opening some interesting issues about the future of “the core”, and probably even more interesting issues regarding networking overall. While the focus of media coverage has been 100G Ethernet, the real question is how networks are valued, or made valuable.
We might call the current situation in the network core a contrast in speeds. Optical fiber is used for virtually all important transport applications, and the state of the optical art sets a per-fiber capacity. To get the most bang for your fiber buck, you’d like to use that capacity, obviously. At the same time, the electrical interfaces in networking have their own capacity level. You’d want to use that too, but the big question in networking is the ratio between these speeds. Right now, for example, the realizable capacity of a fiber strand using DWDM is many times the electrical interface speed, such as that of 100GE. That means that to use a fiber effectively you need to optically multiplex multiple electro-optical paths onto a fiber. The greater that multiplier, the more optical work you do and the more dollars are transferred from routing to optics.
Service providers have been pushing to create a more opto-centric core network for the simple reason that it’s less costly. Optical multiplexing of wavelengths is by some carrier measures a fifth or less the cost of accomplishing the same thing through routing. But it’s also less flexible and less agile in the face of an outage. Nevertheless, pressure from operators (like Verizon, who has had an optical core RFI floating for almost a decade) has forced vendors like Cisco and Juniper to come up with tighter coupling between their routing layer and the optical core. Yes, everyone agrees that this will reduce core cost and thus core vendor profit, but since somebody is sure to do what the carriers want, both vendors know they had to go along eventually. It’s a demonstration that the core network is nothing but a bit pump, something that has always been difficult to differentiate and that will eventually become virtually impossible to differentiate. Huawei, of course, is counting on that and looking to enter a suite of low-cost products in the deep-core electro-optical field at the right moment.
It used to be that the unit cost of a fiber bit was high enough that efficient utilization of fiber was the only issue, but now that cost is declining as fiber technology improves. The value of aggregating traffic in electrical devices to efficiently fill transport pipes is declining with it. Under-utilization of fiber might well be a cheaper option even today (as it is in the core) in some metro areas, and that’s likely to be true in about 75 of the US’s 250-odd standard metro statistical areas (the old LATAs) within two years.
If you can’t justify electrical aggregation in a low transport-cost-per-bit world, then what else is there? One answer is “features”, but everybody has long realized that you can’t get customer- or service-aware in the core, or in fact in any aggregated stream. You need to touch at the edge, where you don’t have to sort out traffic to do that. Another answer is to assert that the management processes are somehow more effective in an electrical network, but operators resoundingly reject that—it’s more expensive to manage higher-layer devices.
The net-net here is that it is not possible to defend core routing. But where core routing loses, edge routing wins. Services are what people buy, and so creating them is what creates the network’s value. Services touch networks where you can afford customer touch, which is at the edge. While we always draw networks as layered structures, even when the old OSI model is long obsolete, the factual map of the network would show the “service layer” and “network layer” touching in the edge device, the edge switch or router.
Cloud services in any form, and in fact the service layer in any form, explicitly undermines the core because you can’t make the core anything other than a bit conduit in a service network. Nobody will be able to defend a contrary position even two years from now, I believe. BT’s discussion of vertical-market clouds is a proof point here; if you are going to focus not only on cloud computing but on cloud computing in an industry-segment slant, you clearly are moving to a level of service differentiation that’s beyond what the network can realize. But because a vertical is a company characteristic, a connection to the company is an implicit connection to that vertical, and you can differentiate by industry at the edge. Moral: The future is all about services and their coupling to edge routers. That’s what you need to watch to understand the fortunes of the big network vendors.