The 100-Tbit Router Race

The Defense Advanced Research Projects Agency (DARPA) appears to have funded not one but two projects to develop an optical packet router capable of scaling to 100 Tbit/s.

Last month, it awarded $12.5 million to Bell Labs for a program called Integrated Router Interconnected Spectrally (IRIS) (see Lucent Gets Grant for 100T Router). Agility Communications Inc., the University of California, Santa Cruz, and Lehigh University are also participating.

A second program, called Label Switched Optical Router (LASOR), was announced last week. DARPA has awarded $6.3 million to a consortium of researchers led by the University of California, Santa Barbara (UCSB). Further funding reserved for later stages of the project could bring the total award up to $15.8 million.

The other members of LASOR are Agility (again), Calient Networks Inc., Cisco Systems Inc. (Nasdaq: CSCO), JDS Uniphase Corp. (Nasdaq: JDSU; Toronto: JDU), and Stanford University.

As noted, the goal of both projects is broadly the same: to build an optical packet router that can send and receive 100 terabits of data every second. That's nearly 20 times the capacity of the largest high-end routers, such as Avici Systems Inc.'s (Nasdaq: AVCI; Frankfurt: BVC7) TSR or even Cisco's fabled HFR, which has yet to be released (see HFR, Where Are You?).

The goal appears to be fairly far out, however, and may not be achieved within the lifetime of these projects, which is four years.

"This is early-stage research," says UCSB Professor Dan Blumenthal, who is working on the LASOR program. "It's very challenging but it's great fun, too!"

Both programs are looking initially at the underlying technologies for high-speed routing, and that means a whole lot of effort is going into optical component integration. "Our belief is that you need to handle packets in the optical domain," says Blumenthal. This will require functionalities such as wavelength conversion, optical label switching, and optical buffering.

The focus is on optics because electronics isn't up to the job of shifting such vast amounts of data. "Every time you switch a transistor, it dissipates power, and the faster you switch, the more power you dissipate," Blumenthal explains. An electronic router running at 100 Tbit/s would consume too much power to be feasible.

A device that Blumenthal terms a photon copier is key to the router being developed by the LASOR team. It comprises a tunable laser, an optical pre-amplifier, and a wavelength converter element based on a Mach-Zehnder modulator structure, all integrated on a single chip. Its function is to regenerate the incoming signal, while also converting it to any one of 32 different wavelengths at the output. The photon copier will be used to route packets -- to physically send them to the correct output.

The photon copier device has been under development for several years already. Prototype devices can switch 10-Gbit/s data streams, while consuming 1.5 W of power, according to Blumenthal. There is a snag with the present design, however, which prevents the input and output channels from using the same wavelength. A more complicated design is being explored, with a dual-stage converter that would be totally non-blocking.

Over at IRIS, Bell Labs is pursuing a different approach to optical routing, one focused on micro-electro-mechanical switching (MEMS) devices. Bell Labs researchers were unavailable to comment further on the technologies being developed.

The LASOR and IRIS projects are being funded under DARPA's Data in the Optical Domain Networks (DOD-N) program, and are meant to be complementary rather than competitive. "It is normal practice for DARPA to fund multiple projects under one program to get diversity in approaches and target different aspects of the programmatic goals," says Blumenthal.

Such projects are a throwback to the days when government drove most of the innovation in technology. "We may be seeing that starting again, and it's touching the telecom area," says Stan Lumish, CTO of JDS Uniphase. "We're looking at this as a good learning experience and a way to sharpen our skills."

— Pauline Rigby, Senior Editor, Light Reading

flyingsausage 12/5/2012 | 3:30:55 AM
re: The 100-Tbit Router Race
1 Mega Watts ?? are you sure ?

it's so much, if so would you need to run a nuclear plant to power it, install it in Alaska for the cooling, and build a 10 meters high cable floor for the cables ?

I don't want to be in charge of cabling any 80 racks telco system :) "Can you please check cable n-¦54243826502 ?" :)

Anyway, if Cisco is claiming this, they'll be able to do it, maybe not today but tomorrow (a full test qualification for such a config could take several months, no ?)
Tony Li 12/5/2012 | 1:41:59 AM
re: The 100-Tbit Router Race
Given the CRS-1, couldn't Cisco get in on this? It seems like DARPA is again funding research that isn't bleeding edge, it's just beyond where the industry is today. We've seen this movie before...

arpaktos 12/5/2012 | 1:41:56 AM
re: The 100-Tbit Router Race > Given the CRS-1, couldn't Cisco get in on this? It
> seems like DARPA is again funding research that it
> isn't bleeding edge, it's just beyond where the
> industry is today. We've seen this movie before...

Actually, Cisco is on this :)

And BTW, other than in slideware, has anybody
seen a fully populated CRS-1 (all 80 racks)
operating with reasonable performance ? And
what about the required 1Mwatt of power that
could easily be used for a small town ...

Just curious ...
Sign In