[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: comments on draft-py-multi6-gapi-00.txt
On Mon, 10 Feb 2003, Pekka Savola wrote:
> This seems to solve a non-problem, that is, how PI addresses are handled
> inside an ISP or between its peers.
> The real problem this does not seem to take a stance on is (but there is
> some text in section 10, not sure if related), who is advertising the PI
> routes to the default free zone (or upstreams, in any case)? If these are
> not aggregates, would explode.
I think this point was made clear enough but apparently not. In
principle everyone announces all their customer routes to everyone.
From a router perspective it is more efficient to decide at the source
which routes are announced where, but this assumes that a network always
knows where a peer needs this information. This means that if you
monitor the global routing table using some kind of route collector
you'll see a huge routing table. This can work because networks simply
filter out routing information they have no use for at this location. So
a router in London ignores routing information for US destinations.
Packet for those destinations are sent to a router in the US that in
turn has no use for European routes. (Note that this other router
belongs to the same AS or a transit ISP.) In other words: because there
are no aggregates announced to other networks and more specific routes
are ignored to keep the routing tables manageable, networks must
transport the traffic to a place in their network where more specific
information is known.