A Practical Navigator for the Internet Economy

OPTICAL BORDER GATEWAY PROTOCOL IS CA*NET3 EXPERIMENT TO ENABLE PEERING WITHOUT PURCHASE OF TRANSIT THROUGH CARRIER CLOUD

DIRECT CONNECTION OF LAMBDAS COULD REDUCE BACKBONE TRANSIT TRAFFIC ENABLING INTERNET TO COPE WITH GIG E BANDWIDTH EXPLOSION,

pp. 1-9

We interview Bill St Arnaud, Director of Network Projects for Ca*Net3 the world's first optical national R&D network. Arnaud describes the beginnings of a trend that may enable enough ISPs to peer directly with other ISPs so that instead of having to buy transit from core backbones for 80% of their traffic they may only need to buy transit for 20%.

The Optical Border Gateway Protocol (0BGP) is an experimental concept which, at this point, is unproven. The first enabling step is dark fiber and the availability of many dozens of wavelengths such that an ISP can then purchase its own wavelength and use it to connect to an exchange point several thousand miles away if need be. At such an exchange point you will be able to do standard peering. The advantage given you by OBGP is that you can get to that exchange point without having to pay a transit costs to an upstream provider.

As the number of available wavelengths multiplies and the prices those wavelengths comes down, ISPs will find that they are able to buy wavelengths from their networks to dozens and perhaps even hundreds of exchange points some of which will be in other countries and on other continents halfway around the world. This is the first step. It's doable today and is beginning to happen as Williams and others are now selling wavelengths to ISPs.

With OBGP any optical switch in a optical network can be treated as an internet exchange point such that autonomous ISPs can interconnect and peer with each other anywhere along the network. Now this has some profound consequences. If you let users at the edge control the routing and topology of the network through control of the ports on the switch that means the carrier in the middle will be very limited in how it can optimize and manage the wavelength routing in the network.

BGP has an options field and a number of proprietary products have been using that field for their own special purposes. They are proposing to use the options field in BGP to turn networking upside down. Today when you connect an ISP to an upstream ISP, the first thing they do is to install the physical fiber. Then they put in the link layer which can be ATM or SONET. Next they establish IP connectivity and finally BGP connectivity. They are saying let us reverse that whole process.

So first they would establish BGP peering. They would say I want to peer with Gordon Cook. They have your AS number and can start a BGP session where they instruct their router to connect us. It is the router then that establishes the physical connection between us. Right here in Ottawa most of their institutions are going to put in either for four or eight wavelengths. They can set up their wavelengths set up to go to whichever universities they choose. Then if someone says let's set up a BGP session with Gordon Cook, a router will take one of those wavelengths and steer it towards a connection that will peer with your network.

In their Quebec university-owned dark fiber network they will do a proof of concept that will demonstrate these capabilities over the course of the next couple of months. They will put together and test a very simple version of the protocol. They already have a number of industry partners involved in the project who have indicated to them that they're ready to take the concept commercially and go with them into the IETF.

For one time investment as little $10,000 it is possible to get wired with dark fiber that will last for 20 years. The biggest single expense for institutions controlling their own dark fiber is internet transit cost. If OBGP becomes usable the amount of transit needed for purchase should decrease to between a fourth and a fifth of what is necessary to buy today. What works to reduce transit expense for universities will also reduce them for small ISPs that could use wave lengths and OBGP to peer with each other.

Arnaud expects that in the future there will be three parallel networks in existence. In fact he would suggest that policy not to try to converge all telecommunications into IP because doing so will increase the cost of IP. The residential telephone market will continue to provide delivery of telephone by twisted pair. There is nothing wrong with a voice-over copper which in fact works very well. A second network may well be broadcast video over coax which again works very well. There is no really compelling reason to deliver broadcast video over IP. I believe there will be a third network dedicated to only IP and this network can under some circumstances carry voice or video or both.

When asked what must be done to continue the scaling of backbones, Arnaud comments: "There are probably three possibilities. I am not a fan of the MPLS approach. MPLS was a technique designed to cope with shortage of bandwidth. Now with the ability to buy wavelengths of light on fiber bandwidth is not an issue. Consequently micro engineering the network is probably not necessary today. Also it does not look like we can build routers big enough to aggregate all the traffic. We will require optical "cut thru" or "bypass" circuits. The challenge that faces is who controls the cut through circuits - the carrier in the middle or the customer at the edge."

"Now, doing it this way is not building a traditional circuit switched model. Circuit switched models imply that for every flow you must start a new circuit. The service will set up a switched circuit and send a web page to you, and then switch to another circuit and send a web page to someone else. This will not scale. Having direct peering with each other with wavelengths and doing bypass around other ones may achieve the same ends. In the standards bodies, you have three approaches. One is called the overlay and this is basically the circuit switched model. This is being promoted by under the label of ODSI. Another one is called peer networking. This is where all the wavelengths are treated like MPLS tunnels. The third approach is ours. In this we say let BGP be the controlling mechanism and let decisions be made more upon the lines of traditional Internet direct peering relationships."

GIGABIT ETHERNET EMPOWERS FIBER TO HOME

INCREASING AVAILABILITY OF FIBER FROM UTILITIES & CATV ENABLES CREATION OF MICRO TELECOMMUNICATIONS PROVIDERS

WORLD WIDE PACKETS' GIGABIT ETHERNET GEAR TO ENABLE DELIVERY OF VAST BANDWIDTH TO HOME AND SMALL BUSINESS,

pp. 10 - 14, 26

We interview Bernard Daines the CEO of World Wide Packets. Daines explains that the explosion of dark fiber over the past five years has been fueled by fiber networks laid by cable TV companies and by utilities in addition to the more familiar carrier networks. In many place in the US utilities are bringing fiber to existing homes and especially to developments of new homes. The dynamics of DWDM and Gigabit Ethernet mean that the cost of bringing fiber to new homes and of bringing telephone, video services, and internet as well has fallen low enough to enable a homeowner to pay for the services that only a few years ago a wealthy corporation could afford.

World Wide Packets is re-engineering gigabit Ethernet switching equipment for use in fiber to the home environments. It is driving down the cost of such equipment to the point where a hub and spoke distribution system for 100 homes in the form of a central community distribution system and subscriber distribution units for each home would cost a total of about $1200 per home. Their equipment could be used to provide telecommunications services to customers of companies like VDN in Montreal about which we wrote in our last issue.

In addition to the Subscriber and Community Distribution Units World Wide packets provides its customer with a Network Management Unit (NMU)that is basically, a network management software application. Fiber owners or network administrators would use the NMU to provision services to their subscribers, manage network operations, and define service levels. It provides a GUI interface, which can be Java-based or Web-based, to control the CDU and the SDU-and to give the cable or utility customers of World Wide Packets complete end-to-end management of their entire network. In short World Wide Packets is enabling the creation of micro-telecommunication companies that by finding community market niches can use the cost efficiencies of fiber and gigabit Ethernet to compete against the older and far larger and more wealthy but technologically backward ILECs.

DEPLOYMENT OF IPV6 HELD TO BE CONTRARY TO INTERESTS OF ALL BUT LARGEST ISPS

DISCUSSION NOW SEES DEPLOYMENT AS INCREASINGLY LIKELY BUT BECAUSE OF COST NO ONE WANTS TO BE FIRST

pp. 15 - 18

We summarize arguments on IPv6 on IETF and Nanog list from late August and early September. It is pointed out that advocates have alienated ISPs whose help they need.

MORE ICANN FOOTNOTES,

pp. 19-24

We publish documents relevant to the campaigns of Karl Auerbach and Larry Lessig as well as evidence of Ted Byfield's efforts to find out how the code for the membership campaign and elections servers was acquired.

ALCATEL PUMPS 5.12 TERABITS THROUGH ONE FIBER, SYCAMORE'S ODSI SET BACK, p. 9