A CONTENT VERSUS CARRIER PEERING BATTLE IN HIGH STAKES BBN VS. EXODUS DISPUTE

WEB FARM BASED BUSINESS MODEL IS WEAK & MAY LACK INCENTIVE TO ENGINEER TCP CONGESTION -- JOHN CURRAN DISCUSSES BEST EXIT & PAID PEERING pp. 1- 11

Three years ago Exodus was one of the second tier national backbones with peering agreements at the public exchanges. Today it is a publicly held web farm company sitting on a war chest of a March 98 $69 million dollar stock issue and July $200 million dollar bond issue. Led by the venerable Ellen Hancock of IBM and Apple fame, Exodus, with only a temporary peering agreement with BBN, has become a national web farm hosting such huge sites as USA Today and GeoCities.

On July 9 Exodus was informed by BBN of a non renewal of its peering. Instead of quietly renegotiating connectivity behind the scenes, Hancock has taken Exodus' dispute with BBN public -- warning her customers in a letter of August 5th that BBN and not Exodus would bear the most pain as a result of the dispute. From what we can determine after an exhaustive examination, it is Exodus that has erred in calling attention to the weakness of its business model. The Exodus model assumes that, since it is now a highly desired content provider, it should be allowed to maintain its no cost peering agreements with global backbones like BBN despite, according to some estimates, an imbalance of traffic with BBN that is something on the order of sixteen to one.

For the past two weeks BBN has been demonized on the NANOG mail list by Exodus supporters and those who insist that Exodus get free connectivity because it has content that BBN customers want. Exodus reacted as though BBN were the only backbone that was breaking peering with it claiming to have transit free interconnectivity with all other providers. We conclude that at best Exodus may have cost free peering with all other majors. But, even if this were true, Exodus has absolutely no guarantee that it will continue past the expiration of its current peering agreements. In reality we suspect a considerable fudge factor. Certainly a reading of Exodus January 1998 S1 on Edgar makes it clear that Exodus was having to pay for some of its connections. The caveats listed by Exodus in its business model in the January 98 S1 are extensive and make for instructive reading in the context of current round of finger pointing.

We publish an argument by John Levine that, based on the assumption that content is now king, justifies Exodus' position. We also publish a discussion with Michael Dillion and Sean Doran where Sean points out the very heavy burden maladjusted TCP web traffic can impose on BBN's backbone performance. Of course if Exodus has the right to transmit its content at no cost, it also will lack the motivation to carefully engineer the behavior of its TCP congestion windows with BBN.

We conclude our article with a 5,000 word interview with John Curran done last Friday August 21. While John was careful to say nothing explicit about any specific BBN peering situation, he offers the most detailed description of both the industry's and BBN's approach to peering that we have seen.

The principles that he explains are quite basic. First that peering in the Internet has been and remains based on the assumption of symmetry in traffic streams. In the case of BBN there must be a ratio of traffic between peers that is not greater than 2 to 1. As a result BBN has upwards of 50 peers. Second, the assumption in the industry has been that both sender and receiver pay for traffic with each paying about half the cost of meeting in what Einar Stefferud recently referred to as the mythical middle. The problem arises that the mythical middle vanishes when peering with a web farm causes asymmetrical traffic. If BBN has to open up half a T3 of inbound bandwidth for every T1 that Exodus has to install, the burden on BBN becomes especially great when hot potato routing means that it goes immediately onto BBN's backbone. If this did not happen and it were Exodus responsibility to carry the traffic flow to an exit point closest the location of BBN's customers, this could be a way of restoring the symmetry destroyed by the web traffic. [The letter from BBN leaked on the NACNet web site made it clear that BBN and Exodus had been experimenting with this approach known variously as best exit or longest exit routing.]

What is only now being recognized is that free peering relationships with carrier backbones and large web farms engender traffic flows that are so asymmetrical that we move essentially to a receiver pays pricing model for the Internet. To quote Curran: "if we enter a world where the senders don't really pay any incremental costs, you face some huge implications. You end up with a situation where, for example, a sender could decide to send you a large video image when you connect up to his web site. . . And I guess I am just a little bit concerned, if not from a business perspective, from a public policy perspective about the multimedia spam possibilities when senders are not paying any real incremental costs for sending more information." Curran suggests that the industry continue to experiment with best exit routing and with settlements based peering where he defines that the content provider would pay only for the portion of its traffic that fell outside of specified symmetrical boundaries.

IPV6 TO BEGIN DEPLOYMENT NEXT YEAR

LIMITATIONS OF NAT TECHNOLOGY IN IPSEC, IP TELEPHONY AND MOBILE COMPUTING TO FORCE CORPORATE MOVES NUMBER ALLOCATION IS HIERARCHICAL WITH OPERATIONAL IMPROVEMENTS, pp. 12 - 16

In an interview with Jim Bound, who is the IPv6 technical leader for the Compaq (formerly Digital) Unix group, we have created a very broad and detailed overview of the status and functionality of Iv6.

As Bound explains it most of the protocol work is now done, a world wide tunneling application known as the 6Bone is in operation and Unix applications will appear next year along with initial moves to allocate IPv6 addresses. He points out that NAT address translation is a band aide approach that makes the use of IPsec impossible as well as the kind of QoS applications needed for the success of the internet telephony gateway market. Mobile roaming from laptops back into corporate nets also won't work with NAT.

He believes that the urgency for corporations to use these applications will compel them to start to implement IPv6 in their networks next year. While they will have to renumber in transitioning to IPv6, it has been designed to make the process easy by comparison to IPv4. IP numbering allocation has been established in such a way that, by means of top level, middle and local agreggators, routing will be simplified and flaps unable to spill over their aggregated boundaries. Unfortunately the vast number of IPv6 address numbers available will not impact the hierarchical delegation of the aggregators and we may see disputes over number assignments continue.

BBN'S BLUMENTHAL DESCRIBES 1995 - 96 REGIONAL NETWORK INTEGRATION PROCESS -- EXPLAINS DETAILS OF CURRENT OC-192 SONET BUILDOUT, pp. 17 - 21

We interview Steve Blumenthal, Senior Vice President of Network Engineering at GTE Internetworking. He explains the steps that BBN undertook to integrate BARRnet and SURANet with its NEARnet core between 1995 and 1996. Critical to this process was a need to standardize a myriad of systems from customer care and billing, to network operations, maintenance and provisioning across multiple organizations while maintaining the rapid growth of all three. Lessons learned here are clearly relevant to MindSpring, Verio, WorldCom and others acquiring existing ISPs. Blumenthal also talks about GTEI's provisioning of its 24 strands of Qwest fiber. GTE is using Nortel OC 192 SONET opto-electronics to light up the fiber. The Nortel equipment supports 16 DWDM lambdas on each strand. GTEI is beginning to provision OC3 and OC12 circuits from the fiber and are offering ATM and frame relay services as well.

In order to provision high bandwidth circuits for sale to other carriers GTEI has installed SONET throughout. Blumenthal covers other aspects of GTE's planned expenditures of five to six hundred million dollars a year for network upgrades for the next several years. He concludes with and explanation of GETI's provisioning of America Online and its global provisioning capability as the result of its alliance with Equant, the commercial arm of SITA.

DOES IANA DO AN OPEN PROCESS?, pp. 21-22, 24

Jon Postel and almost half the country code TLD administrators got together in early August and tried to create among themselves a group that would become the Names Council and speak for the commercial registries as well. We show that primarily behind this plan are ISOC and the country code administrators who, while charging lucrative fees, pledge to keep their organization beholden to Jon Postel and to support his IANA draft by laws. We list the fees charged by those countries requiring no operational presence for the administration of their domains.

We also note the conclusion of Roeland Meyer on August 22 was even more striking: "Actually, you make a real strong point. If the IANA folk have a discussion list, it is certainly closed to me, as well as, most that I know. It could be argued, by Postel and Co, that they don't have time for such silliness. However, he's not available to anyone else I know either. If he's doing something, and I'm sure he is, why do we all find out after the fact? That kind of blows his credibility vis-a-vis open processes. On reviewing the latest Draft- Postel, this is kind of underlined (during my feedback to the NSI review, which Dan and I both did). Jon Postel doesn't "do" open processes."