Anybody else remember a recent thread arguing that QoS should be kept at the edge for IP?

Well Peter Lothburg (please excuse the spelling) at the last NANOG demonstrated this fact rather well when he showed the effects of congestion on latency for various high speed links. I think he started with T3 then higher and higher up to OC-192. The graphs showed latency low until the links were at progressively higher percentages of 100% full. For links at OC-12 you were looking at links that were well above 90% full before any detectable increase in latency occured. That made the point pretty clear to me. Walt