What will spur data center/cloud operators to implement 100G metro connections? The cost of metro links will have to plummet, and there's good reason to believe it won't take that long.
According to Siddharth Sheth, VP of Networking Interconnect at communications chip vendor Inphi Corp. , "industry targets are that the cost will have to be down into the $2-to-$3 per Gbit/s range. Coherent 100G will move to the metro, but it could still be too costly for cloud operators in this application. While coherent technology will enable metro connections beyond 100km, cloud operators are looking for links of less than 100km at a low cost, and there's the challenge."
Approaches to enabling lower-cost sub-100km coherent connections include stripping the bells and whistles off existing solutions, or finding the right trade-offs between what happens in optics versus what gets done in electronics. There's historical precedence for this type of evolution: SFP+, for example, is a high-volume and popular data center optical connectivity solution, with tens of millions of modules shipped annually.
Looking at 10G client optics, for example, the industry evolved from 300-pin MSA, to XENPAK, to X2, XFP, to SFP+, and in transition the optics matured and solutions were less dependent on the electronics. Sucked out of the optics, electronics moved upstream into the system, where they were integrated into larger CMOS ASICs, driving down the total cost of solutions. Winners along the way were the companies that understood the implications of the transitions, and those that could exploit the DSP-based technologies that enabled them.
"The transitions will be similar with 100G," said Sheth. "Cheap does not mean less, it means more optimal architectural and technology tradeoffs and better design partitioning between electronics and optics. Greater feature complexity in electronics will enable lower-cost optics, which will lead to a much lower-cost solution. Features such as DSPs and ADCs with low latency, [and] FEC technologies that deliver the right gain will help achieve $2-to-$3 per Gbit/s 100G cloud optical links," he said.
The most important component advances for data center use in the context of 100G optical links are: advances in higher order modulation (HOM); DSP and FEC technologies on electronics; greater integration and stacking; and innovative packaging techniques on the optics. Silicon photonics will have a big role in moving towards lower-cost 100G links, as it exploits more mainstream process and packaging flows, and better integrates optics and electronics, driving down costs.
MSA consolidation will also be necessary to avoid market fragmentation. Since cloud operators exist in a closed ecosystem, they don't need to interoperate with the outside world, but can use their volumes to drive certain MSA success. The market can, in effect, decide who wins the 100G cloud optical link battle. And the winning solution, according to Inphi, will be the one that offers a reasonably priced (in terms of $ per Gbit/s) module in a QSFP28 form factor optimized for single-mode fiber connections of 500 meters to 2km, probably with 2016 volume production.
It's no surprise that Inphi, which boosted its annual revenues by 12.6% to $102.7 million in 2013, believes it can meet these criteria, developing products based on its own optical elements (drivers and amplifiers) and electronics (SerDes, CDR, ADC/DSP, FEC).
But of course, it's not the only components company with an eye on the 100G metro market. (See Acacia Communications Ships Industry's First Pluggable Coherent 100G CFP Module.)
— Carolyn Mathas, contributing editor, special to Light Reading