[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Geo pros and cons
On woensdag, apr 2, 2003, at 01:38 Europe/Amsterdam, Tony Li wrote:
| Now obviously it is possible to lease a circuit to a
| remote location
| and connect to "the internet" in that location rather than
| close to home.
Since this happens all too frequently, we must deal with it.
Sure. But I say we deal with it as an exceptional situation that may
require additional resources rather than accept this as regular so
regular mechanisms must be able to fullly support this.
The Internet is not a centrally run function. Instead it grows
organically by the needs of the users. In other words, the links
only show up where they make economic sense.
Being routable makes economic sense.
If the routes are proportional to the number of areas and the areas
are growing, then you again have a rapidly growing routing table.
The routes are proportional to the number of users in an area. I assert
that for areas above a certain minimum geographic size, the number of
users in an area is fairly constant.
As we concluded many years ago: for addressing to scale, it has
to match the topology. If addressing does not match the topology,
then additional information in the form of longer prefixes must be
advertised into the routing subsystem. Ergo, if one chooses geographic
address, one must force only geographically based links.
Aggregation isn't a binary thing. CIDR can aggregate PA address space
but not PI address space. Before CIDR, PI was the norm but now it's
quite rare. With geographical aggregation the same will happen: in the
beginning, aggregation will be limited. But unlike CIDR, you don't have
to renumber to start aggregating: when a new local or regional link
becomes available, you immediately gain aggregation.
Also, the requirement for "geographically based links" doesn't have to
be all that bad. If network A wants to handle the Italy geo region in
Milan, but network B wants to do this in Chicago, they can use a
"geographically based link" to peer between A's Milan router and B's
Chicago router. With something like ATM or MPLS this is easy to
implement.
Anything
else destroys the aggregatability of the address assignment. Since
we, as IETF members, cannot decree where folks will connect, geo
addressing is a nice theorectical goal which is unimplementable.
I agree that geographic aggregation doesn't look all that good in
theory, and that it lacks long term scalability. However, it's still
much, much better than "straight PI" where we know the potential for
aggregation is 0. For geo, we know the potential for aggregation is
less than 1, but at least it's more than 0. So in the absense of better
short-term solutions (unless you include forbidding multihoming as a
short term "solution") and considering the fact that this can be
implemented relatively painlessly, I think it is worth it.