A Practical Navigator for the Internet Economy

New Infrastructure Costs Are Plunging

Sycamore Networks to Enable SONET Elimination; Terabit Methodology May Enable Inexpensive, Very Reliable, & Open Architecture Router Fabrics

New Companies' Technology Forces Attention to Cost and Scaling Impacts of DWDM Applications 18 Months From Now


Sycamore Networks, (Part 1) pp. 1- 7

The transition of optical Internets, from test beds into production networks, will open up opportunities for innovative technology deployment that haven't been seen since layer two switching technologies were applied by the first generation of commercial backbones five years ago.

We interview "Desh" Deshpande who with Daniel Smith is co-founder of Sycamore Networks. Deshpande and Smith also co-founded Cascade Systems in 1990. With Sycamore the will follow the same business model as with Cascade which was first to market with a family of ATM switches which were used by the early commercial ISP backbones get more bang for the buck.

Sycamore will have a nine-month product development cycle. Not having to worry about backward compatibility nor about cannibalizing its own installed base, Sycamore can use optical components culled from industry leaders such as Lucent and Nortel. With these parts, it will make chassis and filters for DWDM-based IP-over-fiber network transport equipment. The Sycamore products will enable users of the new dark fiber to light new strands in the most cost-effective manner by the purchase of optical filters for placement in the appropriate Sycamore provided chassis.

Sycamore's approach will enable those who are provisioning networks by lighting dark fiber to eliminate very expensive SONET multiplexors at the higher levels of the network architecture. It allows IP backbone builders to wait until they have a purchaser of a specified amount of bandwidth from point "a" to point "b" before buying and installing the filters needed to handle it. Consequently Sycamore will introduce a form of just-in-time bandwidth provisioning that will cut the time necessary to fill circuit orders from months to days.

Current network build outs demand total SONET provisioning of the full OC-192 capacity of the fiber regardless of whether any of that capacity has been sold. In one example discussed Deshpande estimates that the owner of the fiber would have to pay only 25 to 30% as much to provision the entire OC-192 over SONET capability. He adds that the owner of such fiber would be able to start receiving income after spending only 7 or 8 per cent the amount currently required by fully provisioned SONET. Deshpande speaks of wanting to move network build outs from the "mainframe mentality" of 10 to 20 year depreciation to the PC mindset of three to five years.

The Sycamore equipment is referred to as providing intelligent optical networking because bandwidth segments can be 'peeled' off at network pops without having to undergo optical to electrical interface conversions and back again as is the case with SONET based fiber nets. As Sycamore adds more software intelligence to its products, it anticipates that lambdas may be used much as virtual circuits are in the ATM world to provide connections between sites of a single customer and for Quality of Service considerations. Sycamore's first product releases at mid year are expected to focus on equipment with ranges of less than 500 kilometers. Follow up releases will focus on 1,000 kilometer and 10,000 kilometer ranges.

Alan Huang and Terabit, (Part 2) pp. 7 - 22

Alan Huang who in the early 80's created the Batcher Banyan switch architecture at Bell Labs, received a patent on November 24, 1998 for the design of what he calls a Meta-Router. We are the first to publish information about his design methodology which uses group theory to create a switching fabric that could serve as the fabric of a terabit router.

Haung's plans hinge on the fate of his new patent: "Scalable Switching Networks" US Patent 5841775. The patent outlines way to build a very efficient a scalable router. Huang points out that: "It can be built out of off-the-shelf routers, such as the Cisco 12000. It can be made fault tolerant via a fractional increase in hardware. It works in a synergistic manner with wavelength division multiplexing. It is also very well matched to the demands of voice and video.'

Huang claims that the connection methodology contained in the patent will allow him to switch the capacity of a given Batcher Banyan, Butterfly or Omega topology network with almost two thirds fewer ordinary Cisco 12000s than would be needed if these routers were arranged as a large router farm. He then shows how the addition of routers that increases his minimum configuration by 25% will give the resulting router farm three-fold more fault tolerance with fewer routers than the more accepted and popular connection methods.

In one example he shows how a network composed of 36 routers at nine locations (4 routers per location) could lose all four routers at a single location without breaking the ability of the remaining portions of the fabric to switch its traffic. He suggests that this reliability will be especially meaningful to telcos who will see it as a less expensive way to get the extra '9's of reliability that can otherwise only be gotten by the very expensive and space consuming proposition of having to double their equipment across their networks.

Huang's design can also function as a methodology, and a consulting service offering to backbone architects both the tool set and the contextual knowledge he has developed in creating and testing the process. He says: "On one level, I am showing you how to re-organize your router farm to increase your efficiency and fault tolerance. On a second level, I am showing you how to build a scalable, fault tolerant router. And on a third level, I am showing you how to build a scalable, fault tolerant, communications infrastructure for data, voice, and video."

Huang claims that he has managed to turn scalability into a boring subject. Building a network becomes a checkbook decision. You want bandwidth? Call up Lucent and buy as many channels of WDM as you can afford. You want routing bandwidth? Call up Cisco or Lucent and stitch together a certain the number of routers to build a meta-router that can handle the bandwidth.

We have appended critiques of Huang's ideas by Mike Trest and Sean Doran. Both these two, and other's with whom we have discussed what Huang is doing, point out that his designs do not fit well with the current topology of Internet backbones that aggregate traffic rather differently.

Furthermore, what Huang has to offer will not save users huge sums on routers within the next 12 months. As Huang states: "My message is for the mid term - two to five years. I am talking about the grand convergence of data, voice, and video. They might not care about voice and video now but it is clear that the telco's and cable companies are thinking about data. They need to begin to do some tests with my design as soon as possible to reassure themselves that it works as advertised. While they are testing, they will see that I am offering them a whole new framework from which to do their planning for the mid and long term."

The availability of multiple lambdas serving as channels brings to mind again the virtual circuit world of ATM. Huang sees a flatter and less aggregated network: "The backbone was useful in the old days when you only had one link to Denver. With 40 WDM channels, you can afford to bypass Denver and directly connect San Francisco to New York with one of these channels. Minimizing the distance traveled by a packet is no longer the proper metric but rather the number of router hops."

To take a different example, 'with 40 channels of WDM on each link, you can afford to directly connect Seattle to Los Angles via Chicago." Huang's commodity-priced, hot-swappable open-architecture routers are perhaps more suitable to such a world than relatively small numbers of very expensive and gigantic terabit routers being constructed today for an assumed continuation of our highly aggregated backbones tomorrow.

He is not worried by issues affecting scalability. "Don't be scared by 80 channels of WDM on each link or 80 Cisco routers at each location. You can build up to that incrementally. You don't have to buy them all in one fell swoop. Every 5 years the cost to performance ratio should drop by a factor of 10. It will get cheaper and cheaper as you build it."

We find it intriguing to see both Sycamore and Terabit talking about similar approaches in the new TCP/IP over glass world. In both cases we have new technologies making possible the consideration of a new planning methodology of inexpensive just-in-time bandwidth provisioning and router fabric building. Again, the pendulum seems to be swinging toward a DWDM powered world where quality of service issues, when needed, can be handled by a new lambda at the transport level rather than a dif-serv like protocol at the IP level.

COOK Report on Internet: State of the Internet in 1998, pp. 23 - 26

We publish our annual State of the Internet Survey. It will serve as the introduction to our annual anthology to appear by Jan 21. The title is IP Insurgency: Internet Infrastructure and the Transformation of Telecomm. As we have done in previous years, we will publish the complete text of this article on our web site when we publish the anthology in about ten days.