TRENDS
By KATHLEEN RICHARDS
With the first 40-Gbit/sec optical transport systems approaching their debut, high-end router technologies can't be far behind, or are they? It's hard to tell. Traditionally, router interfaces have followed optical interfaces in terms of SONET/SDH speeds, from OC-48 (2.5 Gbits/sec) to OC-192 (10 Gbits/sec). Yet, the core router companies have remained quiet about their OC-768c development. While most admit to some activity, no products have been announced, outside of assertions by some vendors that their 10-Gbit/sec products will support infield 40-Gbit/sec upgrades. Analysts say that product introductions and even technology road maps from these companies are at least a year away.
"I think you can assume it is being developed right now by Juniper, Cisco, and probably Avici and Hyperchip as far as next-generation products, but everyone is being very tight-lipped about when," says David Passmore, director of research at the Burton Group (Sterling, VA).
Similar to the technology challenges faced by the optical vendors, there is a lot of development work required on the router front.
"None of the merchant silicon that's available is designed for that speed, and that includes switching fabrics, network processors, and classification chipsets," notes Passmore. "The new ones that have just come out or will be coming out in the next couple of months are designed for OC-192, not OC-768. This forces the vendors to develop their own custom ASICs, for interfaces, switching fabrics, and any network processing that they try and do at that speed, although generally a core router that operates at that speed is not doing a whole lot of packet processing; it's all it can do to forward the packet properly. It is really stretching the limits of what's possible to produce a fast enough switching fabric and provide sufficient I/O connecting the fabric to the interfaces."
"The physics are well understood," says Rob Redford, vice president of marketing for the Internet routing group at Cisco Systems (San Jose, CA). "There are just a lot of issues in building ASICs at the right speed and the right density to keep up with it. Within the 2002 and 2003 time frame, all that technology will be there. It's a matter of doing the systems engineering in putting it all together and making it work. When it does come together, it is going to be phenomenally expensive."
Thus, the time-to-market between OC-192 and OC-768 is likely to be longer than earlier transitions because the technology challenges are so much more difficult. In the meantime, when 40-Gbit/sec systems start to populate networks, vendors may adopt different approaches to scaling their router speeds. A single channel operating at OC-768 multiplexed down into four OC-192s so the routers can feed it is one method. Other approaches to scaling include clustering routers-basically splitting the load between multiple chassis-or using one chassis with multiple switching fabrics operating in parallel to achieve the higher speed. "The problem with all those," says Passmore, "is first of all, you can waste capacity internal to the router in passing packets between switching fabrics. The second problem you can run into is that the packets can get switched out of order."
In addition to the technology hurdles, the economics of OC-768 differ significantly from earlier migrations to higher bit rates. When OC-192 first debuted, carriers were investing heavily in the build-out of their networks. Today, carriers' focus is on lowering the total cost of owning and operating these networks.
"There have been some public pronouncements by carriers that one of the reasons that they are reducing their capital expenses is because they are finished digging trenches and putting in fiber," says Pete Chadwick, vice president of marketing at Avici Systems (North Billerica, MA). "To some extent, the whole question around 40 Gbits/sec is: Are they going to be able to afford to upgrade their backbones to it-is there going to be a compelling event to cause it to be cheaper for them? And it's not just an equipment question, it's a construction question. These guys are really focused on, 'How do I invest my money to drive more services onto my existing infrastructure?'"
Cisco, Avici, and Hyperchip indicate that while they are working on 40-Gbit/sec products, the market isn't pressing them for this technology. Hyperchip is also waiting for off-the-shelf, front-end optical components.
Because of the combination of technical challenges and lack of immediate customer demand, how soon the OC-768c router market will develop remains a big question.