Sean, I am not so sure that policing is impossible. If the peer has machines that can you can do traces from, and choose random test times and deestinations, you should be able to get a long way. This could be faked, but it gets you up to the point where reasonable trust could win. You could put a cheap colo machine in their site under some anonymous name and test that way. It doesn't seem like nearly the problem to me of getting the more specific routes. Your problem of the colo doing nasty things by fooling TCP congention control is true, but I don't see anything specific to colo in this. Not to pick on them, but BBN could do this just as well as the colo (in both cases, you don't own the source machine.) Also, where you enter someone else's network with packets, which was what I was focusing on, doesn't change this. The people doing this are also likely the people trying to freeload on traffic differential as well, IMO. You ask what do you do when something related to this breaks? My simple answer is you fix it. I again don't see how this changes anything very much from the current situation other than possibly adding something else to break. I said in my mail that MEDs won't do it, and you agree. So what are the risks and benefits of provider agreeing to send more specific routes at maybe the POP level to a peer that wants to give them a traffic break? How would you go about doing this? jerry