100G Ethernet

Ciena Pushes 100-Gig

The 100-Gbit/s throwdown continues, as Ciena Corp. (NYSE: CIEN) says it's completed a demo of fully 100-Gbit/s wavelengths.

The demo took place at the SC08 supercomputing show just recently. For those of us who weren't there, Ciena announced the news today. (See Ciena Shows Off 100G.)

The single-wavelength claim sets Ciena apart from some recent efforts, as companies like Infinera Corp. (Nasdaq: INFN) and Nortel Networks Ltd. have focused on using multiple wavelengths to carry 100 Gbit/s of data in total. (See 100-Gig Demo and Verizon Adds Nortel to Its 100G Club.)

Moreover, Ciena says it put its transponder into a slot of its CN 4200 RS platform. This, the company claims, sets it apart from single-wavelength demos by Alcatel-Lucent (NYSE: ALU) and Nokia Networks . (See OFC: Hero Experiments and Verizon Talks GMPLS, 100-Gig.)

In a way, Ciena's demo was the opposite of what Infinera showed at NXTcomm. Infinera took a 100-Gbit/s data stream off an Ixia traffic generator and delivered it across 10 wavelengths carrying 10 Gbit/s apiece.

Ciena used ten 10-Gbit/s Ethernet client feeds and asynchronously multiplexed their traffic together onto one 100-Gbit/s wavelength. Ciena also applied some home-baked forward error correction (FEC), implemented in a field programmable gate array (FPGA) to help the 100-Gbit/s signal travel along fiber suitable for 10-Gbit/s traffic.

The 100-Gbit/s signal was delivered in a wrapper meant to resemble what the eventual OTN standard could look like. "We took an interpretation of how that's evolving and used FPGAs to develop an OTN-like frame," says Dimple Amin, vice president of special projects at Ciena.

Ciena's choice of modulation was dual-polarization differential quadrature phase shift keying (DP-DQPSK, duh).

Unlike some other tests, Ciena didn't use a live network to carry the signals, instead opting to bring its own 80 kilometer coil of fiber. With help from the California Institute of Technology (Caltech) , Ciena kept the demo going for 12 hours.

This wasn't the only 100-Gbit/s talk going on at the show. Representatives of ESnet and Internet2 used the show to talk up the need for practical 100-Gbit/s transmission by 2010, recruiting Infinera, Juniper Networks Inc. (NYSE: JNPR), and Level 3 Communications Inc. (NYSE: LVLT) to the cause. (See Internet2, ESnet Want Their 100-Gig.)

— Dimple Matsumoto, West Coast Editor, Light Reading

nodak 12/5/2012 | 3:25:39 PM
re: Ciena Pushes 100-Gig It may be a single wavelength, but it is still two different data streams. Is this an upgrade from 10 different wavelengths, I do not know. That would depend on the cost per bit, reliabilty, total system capacity and system densities.

pavlovsdog 12/5/2012 | 3:25:38 PM
re: Ciena Pushes 100-Gig Wouldn't "serial" have to wait until there is something that actually sources 100G as a single stream? Thats not likely to come until IEEE works out 100GE standard.

Quite a lot of deployed 40G gear is also effectively a mux of 4x10G clients into a single 40G wave.
nodak 12/5/2012 | 3:25:38 PM
re: Ciena Pushes 100-Gig I was talking about serial on the DWDM side. The modulation technique they are using effectively reduces the line right by splitting the 100G signal. Corvis did something similar with their 10G DWDM TX.

As for client inputs, at least one router company has been bragging about 100G interface.

The client side on this has 10x 10G inputs.
Stevery 12/5/2012 | 3:25:37 PM
re: Ciena Pushes 100-Gig As for client inputs, at least one router company has been bragging about 100G interface.

The client side on this has 10x 10G inputs.

Shhhh. Marketing folks want us to refer to 10x10G WDM as "100G". WDM? That's so 90s.

"Ohhh, but you don't have to manage those extra wavelengths, so it's not WDM." [actual marketing quote]
AutoDog 12/5/2012 | 3:25:35 PM
re: Ciena Pushes 100-Gig Nodak is correct, this demo isn't serial, but it is a single wavelength and that's quite important for a number of reasons. The multi-lamda 100G approach (a-la Infinera and Nortel) may be easier, but have a number of pitfalls.

What's interesting is that Ciena has chosen to demo PM-DQPSK. Perhaps more legit then 10x10 and a relatively easy implementation (you can get DQPSK components today) but it too has serious performance limitations.

No surprise that the OIF is standardizing none of the above. Rather, they have agreed that PM-QPSK is the right approach. This is a true, coherent system (not direct-detection) which is what's needed to meet the real-world requirements of a LH/ULH 100G DWDM system.

Problem is, it's *really* tough to build a coherent receiver. Clearly Ciena & Infinera don't have the technology, skills, and/or time to do this today.

Notrel's current claim-to-fame is that they can do coherent DP-QPSK at 40G. Problem is their solution doesn't scale well to 100G. We'll see if someone else can step up to the plate here.

^Eagle^ 12/5/2012 | 3:25:34 PM
re: Ciena Pushes 100-Gig Autodog,

many interesting posts here. Many of the comments seem to be made by folks not actually in the trenches doing either 40G work or 100G work.

I tend to take my direction for the work that the CARRIERs are publishing. After all, they are the ultimate buyers of the technology. Pretty clear they are pushing for 100G, only debate is the actual timing, not the push from the ultimate end user from what I can see.

I think the next couple of years we will see several interesting developments.

Please note that from the transmitter side, PM-DQPSK and PM-QPSK is pretty similar. Just a matter of how much off set of phase and other parameters, but same basic MZI structure. What is a little different is this was RZ- PM-DQSPK.

also note, that NONE of the systems houses have coherent detector arrays. All of the systems houses and even the subsystem and transponder players buy those key components from one of the few suppliers currently supplying. None of them do their own detectors.

I hear there are more coherent receivers under development in Asia, USA and Europe (CIP's hybrid integration comes to mind, OptoCube's work, arrrghhh (my age and loss of memory is getting to me now...) that other French company doing AWG's, others.

My key point here is that there are several variations on a theme going on in the market, but all are using same basic principles.
My second point is that almost ALL optical architectures mux together lower bit rate signals at some point in the "bit tree" that feeds the link (think 300 pin transponders). After all, there are very few actual streams of data that need all by them selves 10G or more. These backbone trunks are all some kind of multiple stream muxed into a single optical lambda, regardless of it being optical ethernet or SONET/SDH. Only question is if you optically or electrically mux and at what signal (bit rate) level.

The work I have seen on almost all the recent trials is over a single optical wavelength (see the Verizon work). The only difference I see in this is some faster silicon and faster links to high speed servers (those client side links) to feed the 100G pipe with fewer but bigger streams of bits. But, the optical technology is not really different from all the other 100G trials.

It is pretty clear that there will be several variations of DQPSK / QPSK and PM versions of both deployed in the network depending on link type, fiber type, edfa type, etc.

Pretty clear OIF is going to standardize on some form of PM-xQPSK for the line side.

Client side will be several varieties depending on reach and application. Most common will be 4x28G for single mode links from 500m to 10km in various flavors. for less than 500m, who knows.. probably lots of enterprise ideas here... arrays of 850.. etc etc....

in some speciality networks whose primary business case is pipes for other large companies and smaller carrier virtualization (Think L3), there will continue to be the links that are arrays of 10x10, but many carriers don't like this as it really does not help them carry more "bit dense" traffic. There are several good reasons for IP trunks between clusters of big routers to be native higher bit rates instead of being stripped across many 10G lanes. And many 10G lambda's does nothing for more bits per hertz per km of fiber per dollar (measure of true bit rate transport efficiency)

hence the push for a "serial approach" that allows 100G to pass through the installed base of EDFA's, mux / demuxes, etc.

Therefore, it looks pretty clear that the first standard to be approved will be some sort of PM-xQPSK with each system vendor taking their own approach (for instance on the coherent receiver do you do homodyne or heterodyne optical signal processing?) to put their own spin on it. And that all of these "spins" will be limited by the kind of fundamental optics that are available from the component guys, both electrical and optical. Given the limited R&D over the last several years, this will be interesting to watch.


Side note: In some places, I am seeing the same equivalent push in metro for component and transponder vendors to come up with ways to send 10G and 40G over the pass band of installed metro DWDM platforms that were originally designed for 32 or 40 lambdas at 2.5g!
AutoDog 12/5/2012 | 3:25:32 PM
re: Ciena Pushes 100-Gig Sailboat,

I agree many of your statements, but there are a few that I feel compelled to comment on:

1) Fundamentally the LH//ULH carriers want to reuse their existing line systems and link engineering rules (which were designed for 10G) and use them at 100G. The same requirements were applied to 40G systems not long ago. The problem is, direct-detection schemes like PM-DQPSK don't get you there @ 100G... They simply doesn't have the reach necessary to cut it.

2) Members of OIF recognized this and definitively decided to standardize on PM-QPSK and not PM-DQPSK for the reasons stated above.

3) The key enabler for DP-QPSK is a coherent receiver which consists of multiple photoreceivers, ultra-high-speed ADCs, and huge electronic adaptive filters. In short, an electo-optical modem operating at ridiculously-high data rates.

A true 100G coherent receiver simply doesn't exist today on the open market. Nortel is closest with its well-publicized 40G coherent receiver ASIC, but it doesn't scale to 100G on a single wavelength. As such, the barrier to entry here is quite high and we'll continue to see demos like this which appear impressive on the surface but won't meet the carrier's fundamental requirements to deploy.

^Eagle^ 12/5/2012 | 3:25:30 PM
re: Ciena Pushes 100-Gig Autodog,

I do not disagree with you. to clarify:

1) yes, that is what I was saying, you need coherent detection on the receive end. However, you can do coherent with either PM-DQPSK or PM-QPSK. What does not get there is DPSK.

2) agreed.

3) we have a difference here in nomenclature. You say there is no coherent receiver on the open market. However, your definition and mine are different. Your definition includes all the parts to make the complete receive section including all electronics and firmware for processing and the asics. My definition is the optical bits. The optical bits are available on the open market. What does not exist is a standard commercial chip set. Therefore, in my opinion, the key differentiators between systems houses in the 100G space will be their electronics and ASIC efforts. Those who will emerge the leaders in 100G probably already have asic programs under development. I agree, the barriers are high. I like it that way. If it were easy, I probably would not have a job. I think on the line side, very few players have real substantive ASIC programs underway. Probably less than 5-6 systems houses. And as you pointed out, there is no commercially available "off the shelf" solution for the electronics nor for the compensation firmware.

lightreadingusername 12/5/2012 | 3:25:10 PM
re: Ciena Pushes 100-Gig
I am a researcher, interested in the requirements for the high-speed side. From the electronics side, can someone tell me:

1) what is requirement at 25Gbps? i.e. 4 x 25Gbps transceivers;
power (full 4 transceivers), frequency loss @ Nquist driving the optimal modulators/receivers?

2) what is requirement for the DQPSK modulation? For 40Gbps, nortel is running a 20GSs, 6b, 1.5W ADC. For 100Gbps, this would require a 6b, 50GSs, ADC, which currently does not exist and I also believe would be EXTREMELY difficult to build.

Sign In