At 12:20 PM 4/30/2003, Stephen Sprunk wrote:
Thus spake "Keith Moore" <moore(_at_)cs(_dot_)utk(_dot_)edu>
> > Today, dial-up concentrators usually have an address
> > range that it used to assign addresses to people that dial in. That
> > means at most a handful of routes per dial-up concentrator in the
> > interior routing protocol. If everyone has their own /48, that means
> > a route in the IGP for each customer that's online. There are no
> > hard and fast rules about how many routes you can have in an
> > IGP, but somewhere between 10k and 1M you run into trouble.
>
> this is an interesting point, but I think it has more to do with whether
> the prefixes are statically bound to customers than the length of those
> prefixes. why would giving customers static /64s result in fewer
> routes in your IGP than giving them static /48s? in neither case is
> there a direct correspondence between the customer's address and
> the concentrator.
IMHO, dialup is a bad example because static IPs per customer are rare;
let's switch to the cable/dsl market.
Standard practice is to connect all customers in a given area (or signed up
in a given period) to a single concentrator via some sort of virtual circuit
(PPPoE, ATM, FR, etc). This concentrator then internally bridges all of
these virtual circuits into a single subnet with a single prefix, giving you
one route for N customers. OTOH, if you assign a prefix to each customer,
you then have between N+1 and 2N routes for N customers. The latter might
be justified if we're truly committed to eliminating NATs, but it costs a
lot more in routes, in administration, and in address waste (assigning a /48
to what is, in nearly all cases, 1-4 hosts).
If standard practice is to give everyone a /48, then the routing decision
at the access concentrator could be programmed to deal with this, ignore
the low order bits, and send the packets on to the proper customer. Given
the decision to "waste" the low order bits on MAC address or other locally
defined values, there's no value in an access concentrator worrying about
that part of the address. The residential gateway would worry about it.
Sounds more than a bit like ISDN sub addresses, where the sub address is
only of interest to the CPE. Seems like a small matter of programming, not
a reason to load up routing tables.