[arin-ppml] Does Moore's law help with routing table growth?
On 12/26/2009 5:13 PM, Michel Py wrote:
>>> Michel Py wrote:
>>> I must have missed a step; how does multihoming with
>>> private ASNs work?
>> Scott Leibrand wrote:
>> Our mutual customer runs BGP with both of us, uses a private ASN to
>> run BGP, and announces both of us his route. We both implement the
>> remote-private-as command (or equivalent) to strip his private ASN
>> from the path before announcing it to our providers and peers. As a
>> result, the BGP table contains the route with two different origin
>> ASNs: mine and yours. That's ugly, but it doesn't really break
> Oh. Well while in that mood we might as well announce PA prefixes, it
> doesn't break anything either. No ASN, no PI prefix. Simpler. We don't
> need BGP either, there's always static routes :-D
PI multihoming could be done without BGP, but you'd want a dynamic
routing protocol, and you'd need to convince your upstream to
redistribute the route into BGP. Static routes, of course, are
insensitive to certain outage conditions, so they're less ideal than,
for example, RIP. In any event, I'm not saying running an inferior
routing protocol, or BGP with private ASNs, is a good idea: just that
it's the kind of thing people will do if you start charging for ASNs.
But to your PA vs. PI point, there's a strong driver towards getting PI
space for any org that doesn't want to be tied to their provider and be
forced to renumber. Since the substitute for PI space (PA space) isn't
as good as the substitute for public ASNs (private ASNs), the demand for
PI space is less elastic (more resistant to price pressure) than the
demand for ASNs.
<Snip a bunch of stuff we agree on>