The market for 40 Gbit/s optical transport equipment is growing strongly and looks set to enjoy a five-year period of deployment opportunity before the 100 Gbit/s market gets into full swing, according to industry analysts.

The latest figures from research firm Dell’Oro indicate that worldwide revenues for 40 Gbit/s equipment are expected grow at a combined annual growth rate of 35%, reaching a market size of $1.2 billion by 2013.

“While the overall worldwide optical market declined 20% sequentially and 11% year-over-year mainly due to the effects of the global economy, the 40G market has shown continued growth and strength with a 1Q09 increase of 8% sequentially and 400% year-over-year,” said Jimmy Yu, optical market analyst, Dell’Oro.

And about time too. Early 40 Gbit/s implementations first appeared in 1999, some 10 years ago, and the technology has been through four product generations. In contrast, 100 Gbit/s technology is enjoying significant operator and vendor interest even though it is still in its first generation.

Almost every major equipment vendor has carried out a 100 Gbit/s field trial in the last year or so. And US carrier AT&T has stressed the need for higher capacity links on its network, stating that “every 10 Gbit/s lambda deployed today will become a 100 Gbit/s lambda by 2012″.

Naturally, this kind of optimism provokes questions about whether carriers might skip 40 Gbit/s and move straight to 100 Gbit/s. However, the analysts contacted by fibresystems.org think that isn’t likely to happen.

“First deployments [of 100G] will definitely be in 2012, as some equipment will be available by end of this year, and some carriers will deploy it just to be ‘first’,” said Eve Griliches, program director for IDC’s telecoms research. “But my guess is that real volume deployment will be in a third generation development of 100G, which will not be in 2012, but a year or two after that.”

David Dunphy, principle analyst with Telecoms Strategy Partners, believes the window of opportunity for 40 Gbit/s technology could last even longer. “40G has a strong life left, and will continue to grow and enjoy good opportunities for the next five to six years,” he contends.

“The challenges in successfully commercializing 100G as well as overcoming the technical challenges are greatly underestimated,” he added. “Commercially, we think it could take at least another three years after 2011 for 100G to fully ‘prove in’ from the business case perspective.”

Griliches believes the fact that the majority of vendors are coordinating their efforts around a single modulation format will help speed 100 Gbit/s to market. “What 100G has going for it is the ability to leverage all the technological development that has gone into 40G to date, which leave less of a new burden on that technology,” she said.

But she agrees that the industry is still very much in the early stages of understanding 100G technology, and whether it will meet the distance requirements set for 10G and 40G — the implication being that ongoing R&D might turn up a better solution than the one currently being commercialized.

“If new modulation formats come to market that make 100G even more compelling, that will push the roadmap out again, leaving the window for 40G open even longer,” she said.

Reproduced with permission. © Institute of Physics and IOP Publishing Ltd.

Although network operators are keen to deploy 40 Gbit/s wavelengths to upgrade their capacity on terrestrial routes, submarine cable systems have stayed at 10 Gbit/s because of the distances involved. But tomorrow in a post-deadline paper to be presented at the European Conference on Optical Communications (ECOC), researchers will describe an experiment that demonstrates the feasibility of transmitting 40 Gbit/s traffic over transoceanic distances.

The experiment involved sending 81 channels at 40 Gbit/s over a distance of 11,520 km — setting a new distance-capacity record.

Key to the breakthrough was a new design of optical fibre, from Amsterdam-based fibre and cable maker Draka. Called LongLine, the fibre has a large effective area, which helps to reduce the effect of non-linearities. In turn, this enables higher launch powers to be used, so the signal can go further before it disappears in the noise.

Large effective area fibre is not a new concept. In fact, Corning unveiled the first large effective area fibre, trademarked LEAF, almost exactly 10 years ago. But to put things in perspective, LEAF has an effective area of about 72 ?m2, while LongLine has an effective area of 120 ?m2 — almost twice the effective area of standard non-zero dispersion shifted fibre (NZ-DSF).

Typically as the mode size is increased in a fibre, the optical field spreads further into the cladding, which leads to higher attenuation and bending loss. To overcome this, Draka applied the technology from its bend-insensitive fibres, using a “trench” in the index profile of the fibre to keep the light contained.

“You can control the bend performance, or you can control the size of the effective area without losing light,” explains Alain Bertaina, marketing director, optical fibre, for Draka.

The other challenge in designing the fibre was keeping the fibre characteristics compatible with negative dispersion fibre. Submarine systems typically use alternating lengths of positive and negative dispersion fibre throughout their length to compensate for chormatic dispersion.

The manufacturing process also has a important role to play, says Bertaina. Draka uses chemical vapour deposition, which allows flexible and very precise definition of the refractive index profile of the fibre. “When we target an [index] profile, even a very complex one, we are able to precisely match the profile we target,” he claims.

• UPDATE 25/09/2008 The transmission experiment was carried out by researchers at Alcatel-Lucent Bell Labs, France, and IRISA/INRIA at the Campus Universitaire de Beaulieu, France.

The channels were spaced on a 50 GHz grid, and used polarisation division multiplexed (PDM) binary phase shift keying (BPSK) modulation scheme with coherent detection. More details are provided in the ECOC post-deadline paper Th 3.E.2, which was presented on Thursday 25 September.

Reproduced with permission. © Institute of Physics and IOP Publishing Ltd.

The standardization of higher data rates is vital if Ethernet is to continue as a ubiquitous end-to-end protocol. Pauline Rigby finds out how standards are progressing.

Ethernet has traditionally evolved in multiples of 10, from the first successful commercial version of Ethernet at 10 Mbit/s through Fast Ethernet (100 Mbit/s), Gigabit Ethernet (1 Gbit/s) to 10 Gigabit Ethernet (10 Gbit/s) – the highest speed available today. But the next multiple — 100 Gigabit Ethernet (100 GbE) — hit a speed bump when disagreement arose between different interests within the Institute of Electrical and Electronic Engineers (IEEE) Higher Speed Study Group (HSSG).

Eventually vendors reached an agreement, deciding to add 40 Gbit/s data rate to the proposed standard for 100 GbE, which allowed things to move on to the next stage. On 5 December 2007, the HSSG officially became the IEEE P802.3ba Task Force, doing the technical work to define the standard. Here Pauline Rigby talks to John D’Ambrosia, chair of the Task Force, whose full-time job is to guide the process and ensure the work finishes on time.

PR: What’s driving the demand for higher data rates?
JD’A: Quite simply traffic was growing at such a rate in network aggregation applications, especially with the proliferation of 10 GbE, that it needed a higher-speed solution. People were talking about aggregating 16 or 32 links of 10 GbE to handle all of the servers they were using. Link aggregation was becoming very unwieldy; we needed something faster.

Naturally the Ethernet switch vendors were the most vocal about the need for 100 GbE, but it wasn’t just them. One of the big differences between this project and its predecessors was the involvement with the end-user community. We had people from Google, EDS, Microsoft Networks and AMS-IX [Amsterdam Internet Exchange]. We had content providers coming in because video drives the bandwidth through the roof.

One of the examples that really blew people away was from Yahoo! Asia Pacific. They were offering a service for streaming classic major-league baseball games, and filled a 40 Gbit/s pipe. They don’t know what the capacity request actually was because they had to cap the service at 40 Gbit/s.

I was talking with an individual who was at Google at the time of the HSSG, and he said that: “People point at Google as the anomaly and say well that’s just Google. I can tell you that where Google was a year and a half ago is where all of the other data centres are coming to now, they’re all starting to hit these problems.”

If the demand for bandwidth is growing so rapidly, how did the need for an intermediate rate at 40 Gbit/s arise?
There’s one chart that really helps to explain this question (figure 1). This is a case where a picture is worth a thousand words. This chart was based on input from a server vendor. He was talking about Moore’s Law and how the server I/O [input/output] doubles every 24 months. When the growth curve for network aggregation was added, what we see is two different growth rates for two different applications. Core networking doubles every 18 months, server I/O doubles every 24 months. When you plot this out over time you can see this leads to a huge difference.

We mapped different Ethernet standards across the lines, and looked at how that relates to the needs. You can see why Gigabit Ethernet was pretty much an instant success when it was released in 1998. It instantly hit the requirements for network aggregation, and a couple of years later it hit the servers.

Computing and networking bandwidth needs for 40 and 100?GbE.

In general people say Gigabit Ethernet was a huge success, but when you ask about 10 GbE, it’s a different story. If you were talking to someone who was judging 10 GbE by the value it brought to networking, those individuals say it was a raging success, and you can see that from the chart. However, if you were talking to people who were judging by ports shipped, i.e. server vendors, they say they don’t really need it and won’t need it until around 2010.

To move forward within the IEEE you need to have 75% consensus. Considering the two different backgrounds of the people present in the room, you can imagine the difficulty the group at that time had in coming to grips with this problem so that it could move ahead.

You must have been very happy that a consensus was reached. Is it unusual to have two different rates being developed for a single standard?
Yes it is unusual; it’s the first time history that we’d had two new rates at one time, so saying it’s a big deal is an understatement. But while it was unusual to do the two rates, when you look at the chart it becomes apparent that two rates was the right decision.
Some people are saying that 100 GbE is going to be late. Is that the case?

There are many people that will tell you that 100 GbE is going to be late, and if you look at the chart you can understand why they think that. When you look at 100 GbE in relation to the networking applications, it just barely meets the curve. We’re going to finish the specification at around the time the need is there, so there’s going to be a lot of pressure on the group to produce this thing as fast as we can. On the other hand, if you look at the relation to the computing applications, 100 GbE will be seven to eight years early.

Is there a risk that by trying to please everyone you end up pleasing no-one?
Let me answer that from the standards approach. There was essentially a one meeting delay before we were able to become a task force, which impacted on the schedule. However, as the chair my primary job is to make sure that a standard gets done in the proper time frame, and as a result we’ve scheduled accordingly. It really hasn’t been an issue since we made a decision to do it this way.

In fact, we’ve come up with a unique architecture that will satisfy both rates at the same time, and we’ve adopted most of the baseline proposals around that architecture. There are a few more pieces that we need to fill in, but we are making incredible progress.

That’s interesting. Tell us how the architecture can support both rates.
Several of our key objectives concern the physical layer specifications, or PMD [physical medium dependent]. Right now the solutions that are being considered are based on ganging four lanes or 10 lanes: four lanes of 10 Gbit/s, 10 lanes of 10 Gbit/s or four lanes of 25 Gbit/s. Those lanes could be physical lanes or they could be wavelengths. There are also some individuals from a transport background who proposed 40 Gbit/s serial, and others who suggested two lanes.

The next layer in the structure is the PMA, which means physical medium attachment, and that’s where the multiplexing happens. We came up with a unique approach that allows us to adjust the multiplexing to support 10 lanes, four lanes, two lanes — whatever we need.

One of the main areas we had a lot of debate on was in the physical coding. We have just adopted an approach called MLD — multilane distribution — which is based on the standard 64B/66B encoding that we do now in Ethernet. Simply put, it’s a mechanism to distribute the data across those different lanes. So now we have a lane-distribution mechanism that carves up the signal into the right number of lanes for whatever PMD we’re going to target.

The task force met in May. How did that move things along?
The May meeting was phenomenal, and I attribute that to the hard work that the task force is doing in terms of the consensus building and due diligence with the technical issues. We actually wound up making quite a few baseline proposal decisions at that point, which is ahead of schedule.

On the multimode fibre objective for 40 GbE and 100 GbE we’ve adopted a proposal that’s n × 10.3125 Gbit/s across parallel fibres. For the 40 km 100 GbE and the 10 km by 100 GbE we’ve adopted a 4 × 25 Gbit/s architecture. The decision on 40 GbE over 10 km hasn’t been made yet.

There are two proposals out at this point — one is for 40 Gbit/s serial and the second is for a 4 × 10 Gbit/s CWDM [coarse wavelength division multiplexed] approach. That will be one of the key decisions we’ll need to make at the July meeting so that we can move ahead.

What’s the timetable for completing the 40/100 GbE standard?
To keep to our schedule we aim to generate draft 1.0 after the September 2008 meeting. The next phase will be a Working Group Ballot — that’s where you open it up to the 802.3 Ethernet Working Group for comment, which is scheduled for the March meeting in 2009. After that comes the Sponsor Ballot in November 2009, where you open it up to the members of the IEEE Standards Association. The standards release is scheduled for June 2010.

For more information visit www.ieee802.org/3/ba/.

Reproduced with permission. © Institute of Physics and IOP Publishing Ltd.