Dare We Aim for Terabit Ethernet?
The next speed grade of Ethernet is only at the beer-and-pizza stage, but discussion is already heating up over whether the jump to Terabit Ethernet should be a consideration.
That would represent a tenfold increase over 100-Gbit/s Ethernet, continuing Ethernet's tradition of leaping forward by factors of 10.
But Terabit territory is so far off that some Ethernet experts think 400 Gbit/s is a more pragmatic goal.
"When I try to stretch out my imagination to how we would do Terabit, I don't see an obvious road," says John D'Ambrosia, who currently works at Force10 Networks Inc. , but who has been a figurehead in high-speed Ethernet standards ever since "high speed" meant 10 Gbit/s. (His comments for this story are personal observations that don't necessarily reflect Force10's plans; Force10 is letting D'Ambrosia consult for Ethernet standards efforts.)
"With all of these projects, whether it's been 10 Gbit/s or 40 Gbit/s or 100 Gbit/s, all these discussions go on in the back room," says The Ethernet Alliance director Brad Booth. "Even in Geneva, we'll sit down for a beer and talk about: What is possible? What if we start too early or wait too long?"
"Geneva" refers to a joint Institute of Electrical and Electronics Engineers Inc. (IEEE) and International Telecommunication Union, Standardization Sector (ITU-T) workshop being held there on May 28, when all this chatter could start turning formal. The agenda tentatively includes something called "Beyond 100G," an ITU-T spokesman writes in an email.
Prior to that, on March 22, there's an OFC/NFOEC morning workshop titled "1Tbit/s Ethernet Transport -- Why, When, and How?," where "proponents and dissenters" are encouraged to contribute ideas.
The argument for 400 Gbit/s is that it's clearly reachable. Not all the technology exists yet, but key elements are in the works: 25 Gbit/s optics and electronics, for instance. Speaking at an Ethernet Alliance event in September, Finisar Corp. (Nasdaq: FNSR) director of engineering Chris Cole outlined how 25-Gbit/s componentry could feasibly bring 400-Gbit/s Ethernet into being. (See The 400-Gig Vision.)
Terabit, on the other hand, sounds harder to develop. Remember Bob Metcalfe, the Father of Ethernet and now a Polaris Venture Partners partner, saying two years ago that Terabit Ethernet might require a rethinking of everything, down to the glass used in fiber-optics? (See Metcalfe Pitches Terabit Ethernet.)
Why talk Terabit at all, then? Because the power users say they don't want to be stung again. (See Facebook: Yes, We Need 100-GigE.)
For example, officials at the Amsterdam Internet Exchange B.V. (AMS-IX) , one of the largest Internet peering points in the world, say they needed 100-Gbit/s Ethernet two years ago and have been in scramble mode while waiting for the equipment to catch up. "We are not going to make the same mistake again," chief executive Job Wittemen tells Light Reading.
The dividing line has cropped up at recent gatherings by the Ethernet Alliance and the Optoelectronics Industry Development Association (OIDA) .
"What you had was those of us who had to build the stuff saying 400 Gbit/s was the more practical leap, while those who were using it said they need Terabit," D'Ambrosia says.
But vendors say they've been stung too, in the form of a prolonged downturn that's extended well beyond the telecom crash circa 2001. Put simply: They're worried that demand for Terabit Ethernet will start in low volumes, and that the high price of early equipment could keep those volumes low. That's not much reward for what promises to be an expensive R&D undertaking.
"If you've got one guy saying he's going to buy 10 ports tomorrow, that's not going to help you," says the Ethernet Alliance's Booth (who used to be at Quake Technologies but says he's now working on a "new project"). Likewise, 100 interested customers won't help if none are willing to pay what's likely to be a hefty price for the first terabit modules.
Vendors are having a tough time gauging the potential hunger for Terabit Ethernet. "At one of the meetings, Google (Nasdaq: GOOG) got up and said they need Terabit Ethernet in 2013. The curve we're using from our charts at 802.3ba [the IEEE group developing the 100-Gbit/s Ethernet standard] says 2015. And someone else said 2020," D'Ambrosia says. "There seems to be a big differential."
Another factor to consider is that component and equipment vendors are still busy with this generation's research. Those that complete 100-Gbit/s devices and interfaces this year will immediately need to work on increasing the density of 100-Gbit/s ports and on lowering the cost. Same for 40 Gbit/s.
So regardless of what speed grade comes next, "the timing is going to be brutal," D'Ambrosia says. "I think I'm going to be a busy boy for the next few years."
The implication here, according to Booth, is that amid all that Ethernet work, the truly important issue right now is in the interconnects. If they're speedy enough, for example, then future generations of optical modules can be kept at a reasonably small size.
That kind of work is better done now, rather than waiting for the Terabit discussion to firm up.
"The discussions that are happening right now are about whether there's something we can do at 100 Gbit/s that would lay down the foundation for the next generation," Booth says.
— Craig Matsumoto, West Coast Editor, Light Reading