Originally from ZDNet News - http://www.zdnet.com/stories/news/0,4586,2673857,00.html

Interview: Vinton Cerf on Private Peering
By Max Smetannikov
Interactive Week
January 15, 2001 4:43AM PT

Worldcom followed in the footsteps of Genuity and Level 3 Communications to publish its private peering policy Jan. 5. Vint Cerf, WorldCom senior vice president, Internet architecture and technolgy, and co-developer of Tranfer Control Protocol/Internet Protocol code, discussed this decision and its potential impact on the Internet industry with Senior writer Max Smetannikov.

Q: How important is this move by UUNet, and what prompted it?

A: First of all, this is a WorldCom move: we have combined and rationalized the UUNet operation. In any case, we think this is consistent with what other parties are doing as well. Genuity and Level 3, for example, have made their peering agreements available publically. We think this is a useful step. I don't know if it will change the landscape [of the industry] but I think it would be helpful. If you are an ISP [Internet Service Provider] and you are wondering if it is worth pursueing a peering relationship with some other ISP, having a public criteria available makes your planning a lot easier, and that would be beneficial for [text unreadable] nature of peering. Some people [text unreadable] and of course, the fact is, it is [text unreadable] which comes a baterring [text unreadable] investments. In paticular, to [text unreadable] people look at this, they will [text unreadable] two parties have to build the [text unreadable] when you have capacity going [text unreadable] [metropolitan area ethetnet] or what have [text unreadable] it helps people understand the [text unreadable]

Q: The companies that establish [text unreadable] Spring and UUNet - are believe [text unreadable] running a club closed to [text unreadable] existing peering practices [text unreadable]

A: I think that is a rather extreme statement. What it really comes down to is what peering is all about. Peeirng is about a bilateral agreement - that means between two companies - to carry each other's traffic without charging, without keeping track of cost. The presumption is that two parties would have evaluated their repsective networks and have concluded that, for the traffic that between them, it is mutually beneficial to enter into this arrangement, this barter agreeement. Otherwise, of course, they will be presumably purchasing transit from each other to pay for it. Peering is a business decision, and many business agreements are not public. When you have a contract with another company, it is not necessarily the case that you put it on a front page of a newspaper. So it doesn't strike me as being peculiar that those individual bilateral agreements were considered private. Putting the criteria in the public record just makes it easier for everybody, but I don't think it prevented effective peering relationships from being established, because many parties were coming to us when I was running InternetMCI, and in each case somebody would sit down and describe what their network did, how big it was, where it went and what traffic was flowing on it, and then the two parties would try to figure out if there was a mutually beneficial barter relationship. Sometimes there was, and sometimes there wasn't. In spite of publishing this on the Web site, we will still discover cases when the party will or will not meet the criteria.

Q: This is already discussion of a large number of small ISPs banding together to meet WorldCom's peering criteria. Do you think people can create associations to meet the peering standard with a conglomeration of networks?

A: I am not so sure that would work that well. Remember, in a peering relationship, two parties are agreeeing to carry traffic to each others customers. And there is a presumption there that there is a single network that connects all of those customers to peering points and visa versa. Imagine now that there is a collection of ISPs that banded together and argue that their aggregate footprint meets the criteria. There is still the question of performance. It's one thing for one party to come to the table and say: "I control and manage the performance of my network, and I can make decisions on how to increase capacity or make changes to the topology." It is a little harder for me to understand what the nature of the barter would be in the event you have a group on the other side that would have an unspecified way of linking their net[work]s together. I wouldn't rule [such a barter] out, but it would be a poitn of real interest and concern for me.

Q: So technically meering the peering criteria doesn't automatically mean becoming a peering partner?

A: Keep in mind that the whole idea behind this is that both parties would mutually benefit from the exchange of traffic. Presumably, so would their respective customers. I would sat that meering the criteria is the first step, and then you have to sit down and see if the exchange of traffic would produce a mutually beneficial result.

Q: How does this peering policy affect Web Hosting companies?

A: There is still a published balance criteria, you'll notice it's a 1.5:1 ratio, which allows some latitude for variation in the flow of traffic between two parties. In the hosting cases, what sometimes one ended up with is one party having a very small scale network, so the traffic moved very small distances and dumped into a larger scale network. So it wasn't 100 percent clear what the mutual benefit was, since the two companies were in two different businesses. Peering structure is primarily intended for two ISPs to carry each other's traffic over roughly a similiar geographic scope and data rates. The companies that are doing Wed hosting are providing a different kind of service. That is not so much a carriage as it is a hosting. I don't know if a peering relationship is the right way to arrange a business relationship in this case. There might be some other rationale for constructing a relationship, but I don't know if barter is the right method.

Q: International network operators have long expressed an interest in peering as a cost-cutting measure to get traffic to Europe and Asia. How would publishing this policy influence relationships with such operators?

A: It won't make much difference. The peering policy is broken into three areas: the U.S., Europe, and Asia Pacific Rim. We don't contemplate peering between entities across the Atlantic or Pacific oceans. At the moment, we plan no barter between these major regions. I don't expect there would be any change.

Q: This is a long shot, but do you see content distribution networks as potential peering partners?

A: I am still scratching my head about content distribution networks, and peering exchanges between CDNs. When you build a backbone of sufficient capacity, it is not 100 percent clear how critical it is to have a distributed content on the edges. if you have enough capacity in the core. So the idea of doing something special to do distribution seems applicable when the bearing network doesn't have enough capacity or speed. I have seen some ideas from companies that may not be visibile yet, that look at it in a somewhat different way. In the cases of a lot of CDNs the idea is to take whatever the source has and replicate it to appropriate places to avoid going across the backbone. But one of the more interesting questions - take video for example - is whether you want to replicate what the source has done in four [or] five encoding methods, or whether you want to take one single distribution format and make it available on the edges and let that edge point do a conversion. That turns out a lot closer to how conventional television works. When you hvae a programming source, that produces a single digital version of the video to be distributed, but when it gets to the cable head, it gets converted by the cable head for tranmission over the cable. If it goes to videotape, it is converted in a different way. So I am wondering if the economics of conversion might become more important than the economics of the transport. Right now, the transport model is where most content delivery networks are focused. I am not an expert in this area, I am just an engineer, but is strikes me that the value-add is what each of these players is trying to acheive, and making copies of things and putting them on the edges of the Net isn't as valuable as doing something else with the content to add value.

Q: But goign to the initial question of peering with CDNs?

A: I am not 100 percent convinced that it is going to be necessary to have these special content distribution mechanisms as we get higher and higher capabilities in the core of the internet.

Q: What does this policy mean for small ISPs?

A: Let's talk a little bit about what happens to a small ISP. There has been some serious misunderstanding about the nature of peering and the beginning of an ISP business. When an ISP is just getting started, it has to deliver connectivity to the entier Internet to its customers. That is what this business is all about. So this ISP is getting started, and then it hears about peering and thinks that would be a cheap way to get connectivity everywhere on the Internet. It turns out it costs you money to build a network that meets peering criteria. And if you decide to peer just at public peering places, you still have to build capacity to those peering places. It costs you money to be a peer. Also as you might remember, customers you reach on a peering basis are just customers of other peers. You can't go from an ISP to a peer to all of their peers a peering relationship. If you did that, the party delivering the your traffic to its customers and to its peers would be carrying all of your traffic with no compensation. Not a good business model for them, so nobody does that. You have to build a significant number of peering relations to reach all of the networks you are peering with in order to acheive full connectivity on the Net. That would be a fairly daunting proposition for most of the start-up ISPs. So they buy transit service from somebody who is committed to delivering traffic to every point on the Internet. And as time goes on, they might start meeting peering criteria for some other ISPs and doing bartering arrangements for some of them. As time goes on, more and mroe traffic will be carried on a peering basis than on a transit basis. I have sen ISPs switching back and forth from peering to buying transit. When I was part of InternetMCI group, we had at least one ISP that used to be a transit customer that then got big enough to pass the peering criteria, so then it became a peer. And that went just fine - we were happy with the arrangment. There are/may be some relationships where peering never makes sense.

Q: The newly published peering policy, is it the same one you had in 1997?

A: In 1997 the peering policy dealt with just the domestic U.S., so we have expanded it. The second thing: The growth of the Internet over the last four years has been quite phenomenal, so we adjusted some of the parameters to account for the size and scale of the Internet as it looks in 2001. So backbones are now OC12 [622 megabits per second] instead of something smaller.

Q: This would create some of the fallout, since some people would say Worldcom is raising the bar again.

A: Keep in mind that this is supposed to be a mutually beneficial arrangement; this is a business decision that we make together with the other party. We believe that a year from now, this criteria will be easily met by most of the peers we have now.

Q: Any final thoughts?

A: I hope we will see similiar actions by other ISP participants.