For roughly a decade, there’s been a growing debate on the balance between optical technology and electrical technology in networks. The optical vendors, notably Ciena, have (not surprisingly) been weighing in on the topic the most, given that they’re likely beneficiaries of a shift toward optical networks. A recent Light Reading piece talks about Ciena’s view. I do think that there are forces operating to shift the focus of operators more to transport optics, but I think they’re a part of a larger potential architectural shift, and we need to explore the big picture before we draw local conclusions.
Networks have always relied on “trunks” and “nodes”, and trunk technology has typically relied on economy of scale in bandwidth/capacity. Fat pipes are cheaper per unit of capacity than thin ones, so there’s a benefit to aggregation. A big part of this is the fact that the cost of a trunk includes a hefty charge for deployment of the physical media, a cost that’s largely independent of capacity.
With fiber optics and dense wavelength-division multiplexing (DWDM) you can create an optical trunk with very high capacity for a relatively small TCO premium over lower-capacity pipe. The trick is to aggregate enough traffic to utilize the capacity. If we assumed a static load on the network, I believe that the opto-electical dynamic wouldn’t be under much pressure, but network load is increasing. Fiber has crept into the access network, but not the kind of fiber the article discusses, fiber that needs IP integration. That requirement exists when fiber is deployed deeper, creating a network topology that uses less electrical-layer handling.
Access networks are the on-ramps to the broader network, and access networks terminate in what used to be called “edge offices”, which in turn link to deeper facilities. Overall, there are roughly twenty thousand aggregation sites in the US and perhaps a hundred thousand globally. Consumer broadband has driven up the traffic level in the access network (wireline and wireless), with video being the major contributor. Higher-capacity access connections mean higher capacity is needed to create trunks at the aggregation points. Aggregation at the edge is typically done in electrical devices, and even if optical trunks are used the devices involved are still routers.
If you really want to see IP/optical convergence, you need to look at what happens behind (deeper than) the edge aggregation. There, the big question is the number of trunks, which if you assume essentially unlimited optical capacity per trunk, depends on how you interconnect the aggregation points. You obviously can’t mesh a hundred thousand aggregation points globally, or even twenty-thousand in the US. If operators are truly interested in IP/optical convergence, then they’re postulating more meshing, and a need to transit through deeper aggregation points with lower cost and latency.
Where this happens and what devices are involved depends on a lot of things, but perhaps most of all it depends on what services drive the network. I’ve been a consistent fan of “metro networking”, meaning the presumption that services will focus on/in a metro center, and I think that metro is both the first driver of IP/optical convergence and the primary battleground where router vendors have to work to manage it.
In a pure connection-driven future, there is no reason to think that IP/optical convergence couldn’t replace all of the core routing and much or all of metro routing. Traffic at any point is just passing through, so smart handling isn’t required. We could expect to see advances in the use of IP/optical interfaces on content caches, and as the unit cost of optical capacity continued to fall, the migration of cache points deeper into the network.
The counter-force to this is obviously non-connection services. If we have to add service intelligence, we need to couple computing to the network, and that’s more easily done with traditional network devices like switches and routers. If those non-connection services focus on the metro, then we probably see traditional IP devices with optical interfaces (the current model) being deployed from the metro outward, which again perpetuates the current model.
So is the whole of IP/optical convergence then a myth? To a degree, yes, if we assume that the convergence really means that optical devices take on a limited IP mission to make connection networks more efficient. I think that even the article hints at that, with a Ciena quote: “61% [of respondents] defined IP/optical convergence as the streamlining of operations across IP and optical functions. To me, that involves multi-layer intelligent software control and automation.” In other words, what operators want is operational efficiency improvements, not convergence of the network equipment. If that’s the case, then there’s little Ciena or other optical vendors can do to move the ball.
The deeper truth here is that the concept of “IP/optical convergence” isn’t what the debate here is really about. We have it in at least one form, the optical interfaces on electrical devices, already. The deep issue, the real debate and the real competition, is over what happens at the metro level. Why? Because if metro doesn’t introduce service intelligence, then simple optical aggregation spreads out from core, through metro, and closer to the edge. We might see a radical reduction in electrical-level (router) devices. If metro does create significant service intelligence, then electrical-level, data-center-integrated, networking concepts spread toward both edge and core, and simple reconfigurable optical add-drop handling is diminished.
Does Ciena see the risk here? Does it realize that if metro services explode, chances are that networking will tend to resemble data-center interconnect (DCI) more than it will optical aggregation? Does it see things like the metaverse as the big potential upward driver of that risk? That’s one big question.
The other one, of course, is what the router vendors see. Operators have been saying since 2012 (to me, and perhaps to others) that their profit-per-bit numbers were eroding dangerously. You can raise them by raising revenues, lowering costs, or both, but obviously if revenues won’t contribute much then cost reduction has to do all the lifting. That means the kind of simplification focus that the article implies. So not only do the router vendors need credible service revenue boosts for their customers, they need to make them happen quickly in the metro, to defend their incumbency against optical encroachment.
Network vendors have been slow to recognize reality here, or at least the electrical-device vendors have. I’m not completely convinced that Ciena is demonstrating that the optical players are seeing the light either. There’s a race for comprehension here, centered in metro and service networking, and the winner is going to be a big winner indeed.