If the core is "well run" (not normally over-utilized) and the endpoints have adequate capacity, then you *can* guarantee the call. (where "guarantee" represents a quality *approaching* 100%, as defined in SLAs...) I assume we're not talking about poorly-run cores here. So what I think you're getting at is, when you don't control both endpoints (i.e., to ensure they have adequate capacity) then you can't make end-to-end guarantees. This is clearly true, in telephone networks as well as packet networks. But it doesn't lessen the value of QoS mechanisms. To reluctantly further the telephone analogy: If all 23 bearers on my PRI are busy I still might want to allow certain sources to complete calls to me, even if that means dropping an existing call. This is a local function that I can guarantee, which benefits end to end communication even if it doesn't guarantee it. And if I coordinate this local function at both endpoints then I'm back to my first statement, that you can guarantee end to end. Are you suggesting that QoS has no value unless it can do more than this? Or am I misunderstanding you? A more interesting question is how to make end-to-end guarantees between endpoints that are on different cores, assuming the endpoints themselves are under a common control. If the provider overrides customer QoS preferences, is this possible? Cheers, -Benson -----Original Message----- From: Hannigan, Martin [mailto:hannigan@verisign.com] Sent: Thursday, 15 December, 2005 16:00 To: Schliesser, Benson; Randy Bush Cc: nanog@merit.edu Subject: RE: The Qos PipeDream [Was: RE: Two Tiered Internet]

Randy-

I don't think your bank analogy is very strong, but never mind that.

I agree with what you're saying in principle, that if a user/customer buys bit delivery at a fixed rate then we should deliver it.

But isn't that the point. You can't guarantee delivery, just as you can't guarantee you won't get a busy signal when you make a call. -M<