Life begins at 40GBit/s

What's that noise? It's a torrent of bits and it's getting louder by the day. A study in 2007 by Nemertes Research predicted that North American Internet usage could rise to around 40 exabytes a month by the end of 2013, based on a doubling of traffic each year from 2008. An exabyte, in case you're not familiar with it, is a billion billion bytes. Next year, global Internet demand will be firmly in the exabyte range, with analysts broadly agreeing that aggregate traffic will surge past 10 exabytes per month.

Philippe Morin, president of metro-Ethernet networks for Nortel, says video will be the main driver for the surge in bandwidth demand, as users become accustomed to being able to download and stream films and TV programmes at increasing levels of definition. This will put pressure on the telcos to upgrade their network backbones, both in the cross-country links and the higher-density metro networks.

Jagdeep Singh, founder and CEO of Infinera, an optical transmission equipment company, says: "Exponential growth in network traffic is pushing telecom operators and ISPs to look at higher capacity networks world-wide. Within three to five years, we will see widespread deployment of 40Gbit/s and 100Gbit/s services around the world."

Everything in modulation

If you think you have heard that before, you have. Almost ten years ago, just before the Internet bubble burst, operators talked confidently about surging bandwidth demand and the need to deploy 40Gbit/s connections in their backbone networks. The collapse of the technology industry after 2000 put paid to that. But now, as the world's fibre lights up, operators are beginning to hit the limits of their predominantly 10Gbit/s backbones.

Rick Dodd, vice president of product marketing for Infinera, says: "40Gbit/s has been talked about for some time. It has taken a while to get into the marketplace, largely because the modulation used at 40Gbit/s is fundamentally different to that of 10Gbit/s. The industry is now on the third generation of 40Gbit/s modulation and we are hoping that it will make it in a broad way."

Modulation may seem an odd barrier to deployment, but the changes to 40Gbit/s networking are a reflection of changes in the way that telcos look at their networks. In contrast to the heady days of 2000, there is now little appetite to bury upgraded fibre in the ground and hook it up to a completely new set of optical switches. Now, the expectation is that operators will pack 10Gbit/s channels more tightly, or run 40Gbit/s and even 100Gbit/s over the same fibre as existing 10Gbit/s connections. Equipment suppliers have been busy trying to show that this is possible.

Verizon and supplier Nokia Siemens Networks described at the recent ECOC conference in Amsterdam a trial of a fibre link that carried a combination of 10Gbit/s, 40Gbit/s and 100Gbit/s channels, each using a different wavelength of light.

Earlier, in March, Nortel provided the equipment that operator Comcast used to demonstrate 100Gbit/s running alongside 10Gbit/s and 40Gbit/s channels at a meeting of the Internet Engineering Task Force (IETF) in Philadelphia. In June, Infinera claimed to be first to take 100Gbit/s into an existing network, deploying an Ethernet service over a link operated by XO Communications between Las Vegas and Los Angeles, a total of about 400 miles.

Keeping in tune

Ten years ago, the idea of running 40Gbit/s alongside 10Gbit/s channels seemed inconceivable. Equipment suppliers have now developed ways to tune the signal so that is possible for a receiver to compensate for inevitable losses of power in the fibre, as well as dispersion. At the same time, the designers of the amplifiers used to boost the signal over long distances have taken the needs of higher channel rates into account. They are using more subtle forms of amplification that control noise better than earlier equipment. These techniques make it possible to squeeze more out of the 'link budget' - the sum of all the fibre losses and amplifier gains. "People want to have the assurance that they can evolve their link budgets without having to re-engineer their networks," says Morin.

Since committees at organisations such as the IEEE are already working on the next step, 100Gbit/s, is it worth making the jump to 40Gbit/s when you could make a slightly bigger leap to 100Gbit/s?

"It's very possible that 40Gbit/s per wavelength may get overtaken by 100Gbit/s per wavelength, since the economics of making a jump to a new wavelength bit rate might end up favouring making a tenfold jump not a fourfold," says Singh.

Mike Capuano, director of marketing for telecom routing systems with Cisco, says of 100Gbit/s carrier-compatible Ethernet: "The IEEE is shooting for a standard on 100Gbit/s to be done around 2010. Although Cisco and other companies are putting a lot of effort into 100G, 40G will still be a fairly robust market."

Transmission vs services

The big question facing operators is perhaps not so much whether to go to 40Gbit/s or its successor but what they are going to send over those channels.

Some kind of packet-based system seems inevitable - when mobile operators move to 4G systems, whether based on LTE or WiMax, the backhaul will be packet-based. This is where the consensus breaks down: the move will take time and operators do not want to rework an entire core network just because packet switching is the currently fashionable technology.

Rick Dodd of Infinera explains: "Services and transmission are very different. Operators want to offer services to their customers. They want an economical way to carry that bandwidth through their network. The two need to be decoupled. They want to offer an arbitrary service without ripping out their existing infrastructure."

Telecom companies are gradually moving towards a packet-based infrastructure, but that is where the commonality ends. Some equipment vendors are clear about what they will promote. Nortel has seen stronger growth in its metro-networking side than in long-haul in recent years and has decided that it could do well out of a sale of its metro-networking side, which is much more packet-focused. Nortel's Morin is unequivocal: "If you are going to bet on a technology, bet on Ethernet."

Cisco is pushing hard on its IP over DWDM (Dense wavelength division multiplexing) approach, which removes even the Ethernet part and simply uses routers to pump IP packets over an optical channel. "This is how to build a next-generation network, by removing a lot of unnecessary equipment."

Infinera's Dodd believes that Ethernet is too young a technology in the operator world to make it into the core network before the 100Gbit/s generation arrives and that operators are likely to stick with conventional SONET/SDH for a while. "40Gbit/s SDH has certainly been talked about for a while. 40Gbit/s Ethernet came out concurrently with 100Gbit/s Ethernet. I can't see 40Gbit/s Ethernet being a service offering: it seems to be more for servers than for [telecom] services."

By the time that 100Gbit/s networks take, operators may start to deploy Ethernet in the core network. If they do it will have to compete with the Optical Transport Network (OTN) protocol, another holdover from the last Internet boom, which is positioned as a transport for all the existing framing formats such as SDH and Ethernet. "You may have a 100Gbit/s Ethernet flavour as well as an OTN flavour of 100Gbit/s," says Dodd.

Even within operators, factions fight for the right to define which technology gets deployed. Take BT's 21CN network. Vendors thought this network would see the UK's largest operator move from the time-division multiplexed (TDM) system of SDH to a packet basis - effectively building an enormous Ethernet network. Then, earlier this year, the company surprised the telecom business by stepping back from that and deciding to pick an approach that would let it stick with SDH kit and add Ethernet selectively: the glue being provided by a switching protocol known as multiprotocol label switching (MPLS).

"The market for packet-based transport is continuing, although there was a bit of a setback from BT," says Morin.

Across the Atlantic, Verizon said over the summer that it would start to use Ethernet to build metro-sized networks, although it is hanging back from a wholesale move away from MPLS. By the time that 100Gbit/s arrives, OTN may provide the glue that operators are looking for. Some may take Morin's bet and opt for an Ethernet-based future. And others may just cut out the middleman and start routing IP packets everywhere, with very little overhead between them and the beams of light upon which they travel.