Re: A new spin on multihoming: multihoming classes.

[Geoff Huston <gih@telstra.net>]

At 9/7/01 01:02 AM, RJ Atkinson wrote:

>         Some few folk think that somewhere between 100K and 300K prefixes
>it isn't stable to have "fairly stable routing which converges
>predictably in reasonable time" when using current BGP and current
>BGP algorithms.  Those folk are credible, but I haven't seen any
>equations yet myself.

Again, we appear to be wandering off topic, but right now its not obvious 
where the scaling issues drive us into instability. While the precise 
number of prefixes is in an of itself not the major issue with scaling the 
routing system, its the 'density' of the topology over which BGP is 
attempting to operate, and the rate of dynamic change of the prefix set and 
associated AS path vector and related prefix attributes which are a 
concern. The suspicion is that these concerns are in some way related to 
the number of prefixes, and that a greater number of prefixes in the 
routing table may infer a greater density of interconnectivity, and an 
increase in the rate of dynamic change that in turn may cause BGP. as 
currently used to take longer to propagate changes and explore more 
'infeasible' routes on its path to convergence.

But as to the assertion that 8,192 is some magical preferred number of 
'root prefixes', then I'm not sure that I can agree.

The leverage which appears to offer some promise still appears to be the 
delineation in the V6 address space of 'identity' and 'routing goop' 
(GSE-styled approach) so that the routing goop can reflect, to the extent 
possible, implicit hierarchies in the inter-provider connectivity space.

(However it is still unclear to me, at any rate, that address hierarchies 
in the routing goop are going to provide 'natural' hierarchical aggregation 
of routing identifiers, but such a discussion is probably still off in the 
weeds with respect to the charter of this working group.)

    Geoff