Here is my sole contribution to this conversation. Christian Kuhtz wrote: | That is, if somebody can figure out a way to do private line (circuit | emulation) through an IP cloud. ATM will be an interim step, in my opinion. Done. Essentially what you do is emulate a connection-oriented network over a connectionless one [jnc bait], and prioritize this emulated connection-oriented service. This entails a design constraint: the aggregate input rate for this emulated CO network to any interface must be less than or equal to the output rate, and traffic shaping must be done to avoid the "bunch up" effect. This also requires a resource reservations mechanism of some nature, ranging from manual configuration coordinated by email/phone to fully dynamic "bandwidth brokerage", and a means of identifying the emulated-circuit traffic at each router where it will receive special treatment in terms of queueing, shaping or both. Please see Van Jacobson, Sally Floyd et al's interesting work on this front at http://www-nrg.ee.lbl.gov/nrg-talks.html and in particular Van Jacobson's presentation "Towards Differentiated Services for the Internet" Also very interesting and related to your question is the research and practical work on Class-Based Queueing at http://www-nrg.ee.lbl.gov/floyd/cbq.html and in particular Dr Floyd's "Notes on CBQ and Guaranteed Service" Van Jacobson also did an excellent NANOG presentation, the video clip of which lives at http://www.nanog.org/mtg-9806/agen0698.html (search down to the RealVideo Broadcast beginning at 10:45 am on Monday, June 8) While you are looking at tasty research notes, please also take a moment to consider the wonderful world of best-efforts services (rather than circuit-emulation) and examine the delicious TCP Friendly page at http://www.psc.edu/networking/tcp_friendly.html and study the paper by Jamshid Mahdavi and Sally Floyd intently. These are among the best things you will read for a while. Please send all your network application developer friends -- especially the multimedia ones -- to that document and to the page in general. The followup paper at http://www-nrg.ee.lbl.gov/floyd/end2end-paper.html and the related links at http://www-nrg.ee.lbl.gov/floyd/tcp_unfriendly.html are also wonderful. Heap scorn on the bad guys, or be prepared to police people's congestion-avoidance. For those of you who don't mind doing some digging and who want to see some practical engineering thinking behind alot of this theoretical stuff, please investigate the IETF diff-serv mailing list. Curtis Villamizar <curtis@ans.net> has made some excellent contributions there (from an operator's perspective) that are worth some study, although many others have had useful things to say. (Many others have not, however, but any operator, or anyone else, who reads NANOG is probably pretty good at scanning for content). Ten minutes with Noel Chiappa [the jnc being baited] is also very worthwhile. If you can't arrange that, I believe Gordon Cook did an interview with him in 1997ish that I remember was well worth reading. Finally, the ESNET people have recently done some proofs-of-concept that the equivalent of CBR can be done across a small-i internet and in principle the big-I Internet in the presence of a mix of traffic. This should encourage those people who maintain that the good bits of ATM -- and yes, even I admit that there are some -- can be subsumed into the Internet if and as actually needed. Sean.