Fifteen years ago, or even 10 years ago, the optical white box would have been a good April Fool’s Day joke. Now, it’s looking like a serious possibility.
Component suppliers at the show told Leopold that “web-scale players have experimented with constructing their own long-haul optical transport systems,” he wrote in a note issued after OFC — which is the annual optical networking gathering, with topics ranging from the business side down to the chemistry of fiber-optical cables.
“Long-haul” means these would be systems that connect the spaces between data centers. They would almost certainly use dense wave-division multiplexing (DWDM), a technology that packs dozens of wavelengths onto one fiber. It’s the stuff of long-range telecom networks.
I’d actually heard inklings about this a couple of years ago. The rumor was that one web-scale company had started asking vendors about DWDM equipment — and once the company learned how DWDM worked, it turned around and started designing its own gear.
When DWDM emerged in the 90s, you couldn’t have done that. Long-haul optics were the purview of analog engineering, calling for eye-of-newt, wing-of-bat recipes.
But today, the optics — the hard part — are squashed into standardized, pluggable, off-the-shelf modules. Moreover, 100 Gb/s optical transport is using digital processing to clean up a signal on the receiving end — meaning the transmitting optics can be cheaper and less precise, the stuff of a white box. It’s not as easy as building a white box server or switch, but it’s feasible.
That’s a danger for the optical sector, where hyperscale data centers are supposed to be their escape route. Telecom (including cable) is a finite market — and shrinking, thanks to big players hell-bent on acquiring each other. Data centers are the expansion vector, even though they’re a niche right now.
Infinera has made it a point to target cloud providers, and this week, it announced it had lit up a European network for Facebook. And as Leopold points out, Ciena‘s second-biggest customer last year, behind AT&T, was a web-scale provider, although it represented less than 10 percent of sales.
The catch is that web-scale players don’t want the nuclear-war-survival features that telecom asks for — and that web players work on time scales that run circles around telecom’s deliberating methods. That’s one reason why Facebook and Google are so willing to build their own equipment.
Even if vendors do appease them, the result would be a customer base that’s even smaller than telecom’s. As the cloud starts to dominate networking — meaning, as Amazon Web Services and its ilk essentially become the corporate network — equipment providers could find themselves chasing the requirements of just a handful of customers.
A hedge is available: the branded white box. Juniper is trying this, putting its Junos software onto a switch from Taiwanese ODM Alpha Networks. It’s a product that you purchase whole, as opposed to a white box that you’d cobble together yourself, but the point is that Juniper is selling only software here, not the hardware. No one would be surprised if Arista Networks did the same thing.
My gut says the branded white box is inevitable in the switching world. The optical white box is a tougher proposition, but it’s become a possibility. It might be something optical vendors will have to investigate.