CIX is the largest MLPA with 38,000 routes.
PCH is the next with about 600 routes.
The AADS MLPA is no longer. The NAP customers voted it out of existence.
MPLA is an alternative to bilateral peering agreements. However, the agreement may or may not reflect how the engineering of peering is down.
The large NSPs don't usually support MLPAs. The small and medium providers usually like MLPAs.
MPLAs are also a way to help the inexperienced ISP to learn about BGP from those who have already done it.
MLPAs do mean that inbound routers have to filter inbound routes. This means that big routers are needed if the number of routes is large.
MLPAs are usually problems for some legal staffs who are concerned about "open-ended" contracts and when all the signatories are not known (and can change over time).
MLPAs can cause an inconsistent peering policy since there may not be MLPAs available at all points at which you peer.
MPLAs can be a problem for providers that don't want to peer with their customers.
There is two clear directions. Some people want to have widespread peering. Some people want private interconnections.
Those who seek private interconnections will have major problems when those fail. Those who seek widespread peering can have interconnection failures and some connectivity is lost, but on a percentage basis, it is no large.
Peering can reduce AS PATH length, and reduce the violatility of routes.
With large numbers of peers, you can make your routing more granular.
Two Goals for this session:
Baltimore NAP: MLPA is not written yet. Expecting about 15 providers, 12 of which have not run BGP before. He is planning to use a route server and needs to come up with a way to insure that folks can't poison each other. The route server can be configured to handle the protection.
O'Dell's Law: The Internet is doubling at some rate. In that period of time, the newbies will be the trainers of the newbies in -- of that rate in the future.
The "community" oriented aspect of PCH is very helpful to the newbie ISP. Does MPLA help in this? Probably not.
What about the "bad neighbor" problem? Bill Manning argues that it is better to peer with folks than to avoid to peering even if they are "bad neighbors" -- Just filter to be sure you can't be hurt.
What about value of peering? People will take absolute disadvantage for along time to get a relative advantage and perhaps to get absolute advantage in the future.