By Casimer deCusatis
One of the best parts of OFC is always the evening “rump sessions”, where you’ll often find a refreshingly uninhibited discussion of the latest technical advances in optics. This past year, the discussion turned to how and when fiber will finally replace copper cable interconnects in data centers. In this month’s blog, we’ll consider some of the recent advances in low cost, high data rate interconnects and what the future might hold for the long-awaited optical data center fabric.
Optical fiber within data centers
Many industry observers feel that optical fiber within data centers is the technology of the future – and that’s where it will always be, somewhere in the future. Fiber offers intrinsic benefits such as higher bandwidth-distance products (which should result in future-proof link installations), resistance to electromagnetic interference, and improved security (due to the difficulty in tapping an optical fiber without being detected). However, the primary justification for using copper at short distances (under a few hundred meters, and sometimes as short as ten meters or less) has been the unbeatable low cost of copper links. For many years, a target of under $1 per Gigabit/second was proposed as a threshold for optics displacing copper at short distances. Many large organizations, including Facebook, Microsoft, and Amazon had proposed similar goals. Although this target was likely not based on any fundamental technology constraints, it was good marketing material, and companies who consumed large volumes of network links found this was a good way to challenge their vendors to create more cost-effective solutions.
A quarter per gigabit
Similarly, at OFC 2017 the industry was challenged yet again to deliver a gigabit/second of optical link capacity for under twenty-five cents. The debate over whether this is achievable, and whether vendors can sustain a business model at this price point, centers around two competing technologies, namely silicon photonics and low power VCSELs. The tradeoffs related to power consumption, manufacturing cost, and other issues have been discussed at length during the past few OFC conferences, and I won’t take the time or space to re-hash those arguments here.
From a business perspective, however, as discussed at the 2017 OIDA meeting some companies seem willing to sell very high bandwidth optics at or below cost in an effort to establish a high volume market for themselves. Such efforts appear to center around 100G PSM4 (Parallel Single Mode 4-lane) technology, which is said to be available recently for under $200 per link. The PSM4 standards are fairly mature, having been discussed in the context of the original 100G PHY proposal for a 500 meter link by the Ethernet alliance task force in 2013. The following year, a consortium including Avago, Brocade, JDSU, Luxtera, Oclaro, and Panduit established a PSM4 multi-source agreement, dedicated to addressing the gap between 100G 10 km single-mode solutions (often using wavelength multiplexing) and 100G short reach multimode solutions based on optical fiber ribbons. Many advocates felt that PSM4 would also provide an economic proving ground for new optical component designs, including silicon photonics, modifications to DFB lasers, and novel approaches such as LISEL (Lens Integrated Surface Emitting Lasers). Today, with 100G PSM4 solutions poised to break the $0.25 per gigabit/second barrier, it’s not clear whether this technology can be sustainably sourced at such a low price point. Development of short reach transceivers remained a hot topic at the last OFC meeting), suggesting that at least some vendors are making a big bet on this approach.
The future of 100G
While the industry has a long history of subsidizing certain technologies in order to establish a new market, it’s also apparent that any high volume introduction of 100G data rates will likely keep the data center industry going for years to come. As recently as 2014 (when the PSM4 consortium was just getting traction), over 90% of the telecom industry (traditional early adopters of high data rates) was still capped at 10 Gbit/second. Further, while Facebook and other large cloud data center operators may be pushing the bandwidth limits of copper links in warehouse-scale data centers, most commercial data centers are still dealing with a lot of 1 to 10 Gbit/second interconnects (and future growth in these areas is likely to take the form of a more incremental jump to 40 Gbit/second rather than directly to 100G). Despite ongoing discussion about exploding bandwidth demands in the cloud and Web 2.0, any cost effective 100G option on the market today will likely satisfy the majority of our bandwidth needs for decades to come. The low cost and complexity advantages of a 4 lane wide solution for 100G might well continue for an extended period of time without vendors feeling pressure to reduce prices further due to a threat from higher capacity optical components. Ironically, the fairly small number of hyperscale data center customers who first set low cost per bandwidth targets as a challenge to the industry may end up funding post-100G data rate technologies themselves, rather than riding on the volumes associated with other markets.
What do you think? Can the current optics vendors afford to keep selling PSM4 at cost, or will this strategy ultimately fail? How long can the industry really live with 100G data rates, and where will the next innovation come from? For a no-nonsense discussion of these issues with the people responsible for designing and building our current optical infrastructure, you won’t want to miss the OFC 2018 rump session discussions. Meanwhile, drop me a line (@Dr_Casimer) on Twitter and let me know what your crystal ball says about the future of the optics industry; I may just use your idea in my next blog.
Don’t miss the OFC 2018 rump session.
Posted: 19 December 2017 by
| with 0 comments