RE: draft-savola-multi6-asn-pi-01.txt

["Christian Huitema" <huitema@windows.microsoft.com>]

> >> If we want to break aggregation in a way that doesn't break
> >> scalability then I think we need to put out an RFC that tells
people
> >> how to do it, so that it's actually usable (no point in doing this
if
> >> network operators filter) and scalable. I think it can be done but
> >> Pekka's draft isn't the way.
> 
> > I think that the issue of breaking aggregation will have to be part
of
> > the general architectural discussion. Just like any other proposal.
> 
> Sure, no problem. But this draft isn't the right way to go about it. I
> think the solution is to be found in the area of shooting holes in
> provider aggregates. But we need to get away from the ad-hoc way this
> is done today which couples a relatively low level of redundancy with
a
> high number of additional entries in the global routing table. 

What the RIR actually implement today is a negotiation between some
customers who want the benefits of "portable addresses" and the provider
community that bears the costs associated with large routing tables.
There is nothing particularly wrong with this kind of negotiation
process. Routing table inflation has a cost, but the cost is not
infinite; portable addresses have benefits, but the benefits are finite.
You would expect a negotiation process to stabilize at a point where
benefits and costs are more or less matched, and you would expect this
point to vary over time.

Obviously, what we do here has an effect. Essentially, we can develop
tools that make portable addresses less attractive, thus moving the
"negotiation point" towards a lesser table inflation. 

I personally believe that multi-addressing has advantages over portable
addresses, because it empowers the host. Selection from multiple
addresses provides an easy way to affect the routing of data in the
network, and thus can provide better reliabilities than the vagaries of
BGP. On the other hand, there are associated costs, e.g. managing
several prefixes in a leaf network, handling egress filtering,
renumbering. 

The bottom line is that our only real way to affect policy is by
lowering the cost and increasing the benefits of multi-addressing.

-- Christian Huitema