On 8 Jan, 2004, at 01:09, Jason Thorpe wrote:

> But we're not talking about N paths, here.  We're talking about one 
> path that can queue the packet multiple ways.

These are idempotent.

What you are suggesting is that aggregation will always hide the 
various queueing behaviours behind a single abstraction boundary you 
call "one path".   Although the past is not always a good predictor of 
the future, the past suggests strongly that this is wildly optimistic; 
moreover the present has ugly examples of deaggregation at L2 (discrete 
MPLS LSPs for specific DSCPs).

I would not be willing to take an apocalypse-started-by-this-approach 
bet, however I also would not be willing to bet that the abstraction 
boundary will remain forever contained within the radius of a single 
hop, or within a single network.

If you really want your mind bent, take a look at the multimetic 
routing literature, such as it is.  Nobody's got a good handle on it.  
(Of course, I want to deploy it operationally a year from now...)

	Sean.

PS - diffserv has become sexy and popular in many enterprise settings; 
for that reason in particular, it probably makes sense to follow Erik's 
suggestion and "upgrade" from TOS-think to diffserv-think in netbsd.