I-D Action:draft-azinger-scalable-addressing-00.txt

Havard Eidnes he at nordu.net
Wed Sep 29 10:51:24 CEST 2010


>> In my book that would instead mean 4 PA prefixes (one per
>> upstream provider, who actually provide you with bit transport
>> for payment), where these more specific routes (typically 4x /48
>> prefixes) are announced to your peers.
>
> Resulting in 4 routes in the DFZ.

Not necessarily.  The extent ot the distribution of your routes
would be into your peers networks.  First off: who's to say those
are part of the DFZ?  With PA-client peers, that would probably
not be the case.  Also, do those routes need to be carried by
every router which makes up the DFZ?  I'd say "obviously not",
that is, unless we're talking about peering between what we still
call "tier-1" networks (which we're obviously not, since we're
debating whether a given network can reasonably be given PA
prefixes from his upstreams, and a "tier-1" network doesn't have
any upstreams, only peers and customers).

> Sorry, but PI is a better solution than that.  ;-)

With PI your route *must* be injected into the DFZ, and it *must*
be carried to every remote corner of the planet which makes up
the DFZ for it retain reasonable usefulness.  Doing PI is like
giving up from the start.

I would claim that there's an engineering trade-off here where
carrying one PI prefix in every router which makes up the DFZ
costs more than having a smallish part of the DFZ carry 4
additional PA prefixes.  Besides: then the decision of whether to
peer and carry those additional prefixes can be a local one,
whereas with a PI prefix, the decision for who needs to carry the
prefix is already made a priori by someone else (the RIR), at the
time of address assignment.

Regards,

- Håvard


More information about the ipv6-ops mailing list