North American Network Operators Group|
Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical
NAP History (was RE: The large ISPs and Peering)
On Thu, 26 July 2001, Curtis Maurand wrote: > A rose by any other name... The fact is, and history shows us, that when > cartels form, things get bad for the consumer. Oil, Electricity, > telecomm. However, The placement of the NAP's is disconcerting, because > the process for choosing them was closed. Does it make sense for all of > my traffic going to maine.rr.com from lamere.net (both in Maine and in the > same communities) to exchange traffic at MAE east 650 miles away? The first selection of the NAP's was an "open" process a number of years ago when a group at NSF choose where to place the NAPs. NSF issued an RFP, there were a number of proposals received, committees of outside experts, notification through public notice, etc. At that time three NAPs were chosen (San Francisco, Chicago, and New York) and one alternate (Washingington DC). Over the last half-decade, and additional public exchange points have been formed, and probably an equal number of private exchange points. Some of the additional exchange points have grown very large, such as CIX, MAE-West, LINX, AMS-IX, even though they didn't have NSF's "stamp of approval." If maine.rr.com and lamere.net want to exchange traffic in Maine, nothing changed. The existance or non-existance of a group of national exchange points doesn't change the argument for or against local exchange points. I still believe it makes sense for local and regional ISPs to have local and regional exchange points. You don't need an OC48 to peer with other ISPs in your region. A couple of years ago, when several large carriers decided that NAPs didn't make sense and they could provision private circuits directly between each other better, faster cheaper; several of us (see the NANOG archives) argued it didn't make economic or engineering sense. I even argued ending the existance of public NAPs raised the barrier to entry for new market entrants because of the "combination problem." For example, if everyone requires three points of interconnection, with common NAPs a new market entrant needs to build out three inter-city locations at the common NAPs. But if five major ISPs each require three points of interconnection, but they each choose a different set of cities for new entrants to connect to their networks, a new entrant may need to build 15 inter-city locations. The IETF holds a common meeting of its working groups in location three times a year. While participants going to a single working group may not find it as efficient than having private meetings, it is more efficient for people who want to attend multiple working group meetings to hold them all at the same location. Airlines have created "hub cities." However, airlines set up their hubs so most of them don't overlap. The other alternative is setting up a common location with most of the major competitors, such as the New York Stock Exchange or the Chicago Board of Trade. Markets can be more efficient if all the participants know they can go to a common location, and compete in that location. Markets seem less efficient when competitors have seperate hub locations. Are there downside risks, of course. When the CIX router moved from the Worldcom Santa Clara POP to the PAIX, at least one major ISP used that change as an opportunity to terminate some of its peering. Did the presence of the CIX router increase the business at the PAIX, probably. But its not a given. LINX has built out the LINX fabric in several locations in London. The presences of the LINX helped some locations, and didn't help other locations. If Telehouse has problems in London, the existance of the LINX magnifies the impact. If the New York Stock Exchange has a problem, it affects a lot more companies than if they each traded on seperate exchanges.