A Practical Navigator for the Internet Economy

Optical IP Backbone Revolution Emerges Canarie Runs IP Over WDM Eliminating SONET and ATM -- Expects Cost Savings of Greater Than 95%

Avici Terabit Packet Switch Capable of Oc-1536 -- Plummeting Costs Mean Creation of "Inexpensive" New Carriers Able to Out Perform Old Giants pp. 1 - 6

A technology cost tsunami is about to break. When the flood recedes, expect the landscape to be drastically different. An interview with Bill St. Arnaud of Canarie explains the optical backbone network project for which he has just received funding. Using either gigabit Ethernet framing or SONET framing but no ATM and no SONET equipment, it will move TCP/IP packets directly over Wave Division Multiplexing. This will gain back a 25% overhead in bandwidth eaten-up by the suddenly archaic transport technologies. But far more important is that the SONET and ATM equipment rendered obsolete may represent a saving of more than 95% of the cost of delivering bandwidth. Bandwitdh that cost more than $5000 a month to deliver in January 1998 may cost the new players as little as $50 a month to deliver before the end of next year. Of course 'cost' does not equate to price.

St. Arnaud points out that he can take an off- the-shelf Cisco GSR 12000 gigabit router a get it to route TCP/IP over WDM. While the network will require no technology breakthroughs, it will require technology integration and work on some of the routing protocols to function with maximum efficiency. It will be used for beta- testing technology from Pluris, Juniper and Avici which on May fifth announced a breath taking device that puts lasers and router in the same box and can reach an amazing OC-1536 (160 gigabits per second) over a single strand of fiber.

Economic implications for the telcos are profound. Some folk expect legacy SONET networks to run parallel to the new optical networks for sometime. Others say the problems that will be raised for the incumbent LECs and IXCs will be so serious (as evidenced by some of Qwest's recent creative deals) that they will have no choice but to begin to cannibalize their legacy networks as quickly as possible. Given that they could quickly find themselves with operating and depreciation costs that are up to ten times higher than the next generation telco's (Qwest, Level 3, etc.), they may be in a serious position rather quickly. Look for them to hold on to the local loop to keep from drowning. The bottom line, as Jack Waters Engineering VP at Level 3 told us in an interview on May 2, is that Level 3 resulted last year when founder James Crowe realized that it was possible to invest between a billion and ten billion dollars and build a state-of-the-art telecommunications company that would have cost $100 billion to build in the early 90's.

Some of the implications near term for the public internet are interesting. As the backbone "x" interview at the end of this issue shows, the big five have private interconnects locked up and open peering at the public exchanges is essentially worthless. (Verio probably the sixth, seventh or eight largest backbone can't get private peering with UUNET.) Therefore those not in the top five get increasing squeezed in costs to the point where they effectively have to become customers of the top five.

Now enter Qwest and Level 3 with national and international fiber nets of their own and the ability to offer almost unlimited backbone capacity to major corporations and second tier NSP backbones. If the pricing is right, they should fill their backbones quickly turning from a large intranet into a shadow internet. They must still buy connectivity to the public internet controlled by the big five in order to be viable. Their low cost of operation should enable them to pay whatever prices are necessary. Furthermore it will be possible to measure customer traffic flow into and out of the public Internet and charge a price differential for that traffic. If the connection price charged to begin with is low enough, customers should not mind paying the differential and the next generation backbones should very quickly become magnets that attract business that would otherwise go to the big five. The key indicator to watch in coming months will be Qwest's and then Level 3's pricing.

Internet Telephony for the Stupid Network Small Canadian Company Produces Device to Connect Phones To IP Nets

Explains an Architectural Mindset & Standards Framework By Means Of Which IP Telephony Can Replace PSTN, pp. 7 - 15

Francois Menard wants to completely redesign the telephone network in a way such that voice inside IP packets with replace the PSTN. He offers users an inexpensive hardware device that functions as a bridge between an ordinary telephone and an IP carrying Ethernet, or cable modem service. While his device will interface with a PC, he seeks to have it function independently of a PC by moving the intelligence of the phone company central office switches to the edge of the network. His phone adapter will have enough intelligence in it to function as a client and will interact with server software located inside of the domain of the service provider. As he puts it, the phone adapter contains that extra small amount of computing power that must be deployed in the end-user's home "if we are to succeed in introducing new services to the network without going to the expense and complexity of employing telephone switches inside the network to achieve them. We are taking the network intelligence that is prominent the central office switches and moving it to the edge where signaling and call initiation will all be done inside the telephone."

He sees this happening in the context of the general demand for faster Internet access. This demand is forcing the telcos and the cable companies to make architectural changes to the networks that will render them capable of providing Internet telephony as an unintended by product. The ability to plug telephone lines into this architecture is a welcome, but almost inadvertent, dividend of the architecture itself. The key to all of this is to be able to apply very low cost telephone appliances to the opportunity. The phone adapter that Mediatrix makes is one such appliance.

He sees the future as one, not of IP telephony toll by pass, achieved by IP based detours around parts of the PSTN, but as one where connectionless IP based networks replace the phone companies circuit switched networks. The new world is one where everything becomes part of a data network and where you cannot expect to open data packets to ascertain whether they are voice or video or ASCII text. You can no longer think about charging according to the kind of data the packet contains. Doing that would be totally inefficient. Therefore you have to focus both on charges for quality of service delivered and new telephony services that can be delivered by intelligence under the users' control at the edges of a big, fat, fast, stupid IP network. By making a device that enables an ordinary phone to reach another ordinary phone via a simple cheap bridge to an IP enabled telephone network or cable TV network, he hopes to encourage both industries to adapt his bridge out of fear of loosing customers to the side which offers services that the first cannot.

Within 12 months he wants to turn his bridge into an IP telephony chip. That he believes will cause the dominoes to begin to fall. In his words: "Once we have an IP telephony chip, then adding IP telephony will not involve a significant cost for anything that speaks IP. At that point the only unanswered question becomes what your dominant home local area network technology will be. Regardless of the answer, you will begin to be able to put new wall-plate, data-speaking jacks all over your house. Either you will do it with the telephone wiring or you will decide to call your local cable vendor."

Finally in talking about standards development, he explains H.323 as the protocol favorite of Intel and Microsoft. H.323 is big and bloated. It is so complex (originating in past from Intel's work on ISDN a few years ago) that companies smaller that Intel and Microsoft have trouble dealing with it. He contrasts H.323 with SIP a smaller and lighter protocol under development at Lucent and elsewhere. When SIP is finished at the end of 1998 he predicts that it will triumph over H.323. H. 323 is a way to standardize a protocol, while SIP is a way to standardize a generic means of conveying session initiations on the Internet. SIP is layered. If you want to do more things you simply add another protocol on top SIP. With the H. 323, if you want new features, you need to build them the inside the ASN.1 structure which is in turn inside of H. 323.

As long as you have the PSTN in the picture -- you will have a problem with Internet Telephony. Gateways are merely a means of interfacing the PSTN with Internet Telephony.

Effort Ramps Up to Map Phone Numbers to DNS Viable Protocol Mapping Sought to Enable IP Telephony to Transparently Cross Provider Boundaries, pp. 16 - 20

An effort is afoot to find a way to map phone numbers from the PSTN to DNS or some naming system that would enable Internet telephone calls to go transparently across provider boundaries and into the PSTN. Discussion with Richard Shockey and the new IPTel mail list.

Peering: Backbone X Gives Most Candid Look Yet

Private Peering Unobtainable -- Public Exchange Full Peering Rendered Worthless by Bandwith Bottleneck Between Exchanges & Big Backbones pp. 21 - 22, 24

On the condition of anonymity, a backbone has given us the most detailed summary of the actually operation of peering we have yet seen. It is likely that the peering agreements between the big five and 30 or so smaller backbones at the public exchanges won't have to be abrogated by the big five. The reason is that, as traffic growth at places like MAE East has doubled in the last year, the big five have not increased the capacity of the pipes from their backbones to the exchanges. These pipes have turned into bottle necks that render the exchanges increasing worthless as a means of interconnection to the big five.

The smaller backbones are being offered variations of peering. Backbone "x" describes "paid peering," and "non shared peering" as way stages set up between the increasingly worthless free peering at public exchanges and fully shared privately interconnected peering. What this means is that the second tier backbones have essentially three choices: maintain increasingly non viable public exchange peering with the big five. Find the money to become paid customers of the big five or switch to Qwest and Level 3 if acceptable pricing and transit is available. An irony is that because of non disclosure agreements it is very difficult to know how a second tier backbone's status is changing since the backbone itself is happy to keep the non disclosure in effect. Doing otherwise would be to let the public know its status had been diminished. We vetted our draft with two other backbones which told us they found it accurate.