Data center networks finally should see the deployment of 100 Gigabit Ethernet optical transceivers in 2016, observers believe. A new generation of cost-effective, high-density optical modules provides the main catalyst for these long-awaited deployments.
The IEEE completed the first round of 100GbE (and 40GbE) specifications in 2010, as IEEE 802.3ba. While 100GbE interfaces have found acceptance as carrier router and client-side interfaces on optical transport equipment, there are several reasons it will have taken six years for data centers to adopt the technology.
"One has been the cost of the optics. The optics have been quite expensive, and the solutions that were initially standardized don't meet the requirements of the data centers," explains Daryl Inniss, practice leader for components at market research firm Ovum. "Five hundred meters to two kilometers is a new distance that the Internet content providers were looking for, and there were no transceivers that supported that distance. The transceivers either supported 100 m or 10 km; 100 m was too short, and 10 km was way too expensive."
A Google-led consortium, the 10x10 MSA, first brought this disconnect to light with an agreement to pursue a transceiver that would support that middle distance between 500 m and 2 km via an optical module based on 10 lanes of 10 Gbps. That MSA failed to catch on in the data center market. But Google and it peers continued to request an alternative to the 100GBASE-LR4 specifications aimed at 10-km applications. Industry responded in 2014 with several MSAs that relied on a variety of technologies, from parallel optics (the PSM4 MSA) to WDM for networks based on single-mode fiber. Chief among the WDM group were the CWDM4 and CLR4 Alliance.
Thus, data center operators suddenly had plenty of options. Which naturally leads to the question of which options will see the most traction.
"It's clear that there are at least two that are gaining quite a bit of traction," Innis says. "First there's the PSM4, the Parallel Single-Mode 4. That one has been championed by Microsoft, so there's definitely a customer pulling that. And then there are the CLR4 and CWDM4. These are two different MSAs, but the optics in the transceivers are the same. And so we anticipate very nice traction from that, and we've heard that Facebook is very interested in that."
"I think from a single-mode point of view the CLR4/CWDM4 will be the most cost-effective," says Adam Carter, COO of Oclaro, by way of amplification. "You don't necessary have to cool the lasers to do that. So you obviously don't need special cooling inside the package; that helps with cost and then it also helps from a yield perspective. So I think you'll see the CWDM and the CLR4 will be in that space for quite some time."
Three major data center operators will drive most of the deployments, according to Vladimir Kozlov, CEO and founder of LightCounting. "I think two of the customers are committed to PSM4 technology. One in particular, Microsoft, was very open about it. I think the second one is less committed and open to multimode as well. And I think Google is planning on using everything, not just multimode but a combination of multimode and single-mode. And that's customer Number 3."
Timing is everything
Having spoken so long about 100GbE in the data center, both analysts and technology developers feel it's about time to get deployments rolling. But while 2016 is right on the horizon, significant deployments aren't likely to begin as soon as the calendar page turns.
Opinions vary about when during the year the deployment wave will start. Kozlov polled his vendor sources and discovered "most of them are saying that it's not going to happen until the second half of 2016. Which is a little bit of a disappointment, because I was hoping for Q1, at least Q2 next year."
But not every vendor is so pessimistic.
"Regarding the inside data center applications, our understanding and read of the market right now indicates that around the middle of next year, beginning in Q2, is when top-of-rack switches will really start to use 100-gig optics," reports Manish Mehta, executive vice president, product line management at Source Photonics. In addition to the PSM4 and CWDM4/CLR4 variants, Mehta sees demand for 100GBASE-LR4 Lite optical transceivers.
He also sees a few behind the scenes elements that could affect the timing of transceiver availability. "I think there's an entire ecosystem that has to come together. ASICs have to be available, memory has to be available, connectors have to be available," Mehta points out. Transceiver vendors should have a clearer picture of the state of these components as 2015 ends, he says.
One area may pose a concern, according to Mehta. The indium phosphide based 25-Gbps DML and EML chips necessary for 100GbE transceivers also will be in demand for other data center applications, as well as fiber to the home and wireless backhaul. "The availability of those chips is something to watch closely as operators and OEMs begin to deploy their networks," he warns.
However, Mehta is confident that suppliers can overcome the potential chip bottleneck, leading to 2016 indeed being the year when 100GbE finally arrives in the data center.
For more information on optical transceivers and suppliers, visit the Lightwave Buyer's Guide.