OFCNFOEC: On to 100G
Martin Rowe, Senior Technical Editor- April 1, 2008
From listening to the talk at OFCNFOEC 2008, it's clear that the fiber pipelines, once partially dark, are now illuminated around the world. This year, there was no more talk about how fiber to the premises (FTTP) is going to save the industry. Emphasis has shifted to core networks to carry the increased network traffic. Today, people are looking toward the next big step in core-network technology: 100-Gbps Ethernet, known here as “100G.” (100G, formally IEEE 802.3ba, was announced in December 2007.)
OFCNFOEC: The increasing demand for bandwidth
OFCNFOEC 2008: On to 100G
The new bottleneck
For the last several years, 40-Gbps fiber links were considered laboratory projects. Now, they've started to move into deployment, as evidenced by the fact that people have started talking about reducing the cost of 40-Gbps optical components.
At the plenary session, Pieter Poll, CTO at Qwest Communications, told the crowd that “IP traffic is increasing faster than Moore's Law” and that 100-Gbps networks will need more complex circuits than 10-Gbps and 40-Gbps links. He said he expects IEEE 802.3ba to be ratified in 2009, with components appearing on the market starting between 2010 and 2012. He also noted that today's 10-Gbps networks need to become more economical and more reliable. He sees 40 Gbps as an interim technology before service providers move to 100 Gbps.
Following Poll, Ethernet co-inventor Bob Metcalfe went a step further, predicting the advent of the Terabit Ethernet and speculating that the first terabit networks in labs will appear in 2015. “Applications will drive Terabit Ethernet,” he said, “and television will top the list.” In fact, you can already see it coming given the proliferation of IPTV and Web sites such as YouTube. “If we build Terabyte Ethernet, they will come,” he said. But he warned that moving to terabit networks will require a new fiber infrastructure, claiming that the existing “plumbing” won't handle the speed.
Metcalfe also predicted that Ethernet will eventually kill SONET because Ethernet standards provide no design options, and thus improve on interoperability of networks. “People fail because they put too much emphasis on standards compliance and not enough emphasis on interoperability.” He went on to say that “OC-768 will be the last SONET standard. “There will be no OC-3072.” (SONET usually leaps in multiples of 4, thus 4x768, or 3072.)
OFCNFOEC featured a panel discussion with members of companies ranging from Facebook to component maker Avago Technologies to Reuters. Each panel member discussed how video-based applications such as Facebook and YouTube that stress core networks are fueling a demand for ever-more network bandwidth. Don Lee, network engineer at Facebook, told the audience that he needs 100-Gbps Ethernet right now and he'll need Terabit Ethernet by 2011. Lee cited Facebook's 66 million active users who load audio and video onto their pages, and he noted that Facebook adds 1 million new users each week.
Dave D'Andrea of Avago Technologies cited a problem with 10-Gbps optical transceivers: too many configurations. “Today's connectors and transceivers never achieved a low enough price,” he said, which keeps costs too high for many data centers to move to 10-Gbps links. “Unlike Gigabit Ethernet and Fast Ethernet, connectors and transceivers for 10-Gbit Ethernet are not being made by the millions needed to get the economies of scale that push prices down.” D'Andrea is pushing a new connector called SPF+ as the solution, even though the standard for the transceiver is not yet ratified.