Re: [RRG] Long term clean-slate only for the RRG?

["William Herrin" <bill@herrin.us>]

On Sat, Jun 28, 2008 at 10:07 AM, David Conrad <drc@virtualized.org> wrote:
> While I agree that a cost/benefit analysis needs to be done, I think it
> worthwhile to keep in mind the alternatives.  The alternatives (as far as I
> am aware) are:
>
> a) PI for everyone

David,

If I understand your meaning here, I'd describe this as "BGP PI for
Everyone => Randy Bush's $10M routers."


> b) NAT
>
> I hope you'll agree that alternative (a) alone is not scalable in the long
> term.  One can imagine a scenario where you have a universe of NATs
> connected via PA assigned end points, however I'd argue this is actually a
> locator/ID split where the IDs are not globally unique.

I disagree with characterizing NAT as an alternative. NAT's likely
contribution to the routing problem is already represented in the
status quo. Unless you're aware of a clever new way to use NAT to
relieve routing pressure that isn't an obvious non-starter?


> Am I missing an alternative?

Two more:

1. Status Quo.

We continue to suppress table growth at the RIR level. Users continue
to spend uncounted manhours on renumbering tasks. Users continue to
lose productivity to address-state issues (e.g. spam filtering). The
indirect cost due to the BGP table size continues to creep northward,
helping squeeze out smaller ISPs. Mobility remains cumbersome. IPv6
without NAT remains nonviable because of the interior server
renumbering problem.


2. New IP layer-4 protocols and change everything up to layer 7.

Suppose that in addition to propagating routes, the routing protocol
also performs optimization of address assignments. That is, it
dynamically instructs "downstream" systems to change their addresses
in a manner that improves address aggregation "upstream."

In a perfect hierarchy you'd get perfect topological aggregation. The
internet is not a perfect hierarchy, so in order to still get good
aggregation you have to draw some line between systems large enough
that they won't accept renumbering requirements from their peers and
systems smaller than that which accomplish multihoming by acquiring
multiple address sets from each upstream hierarchy.

The only way to combine the two effectively is to remove identity from
layer 3 addressing altogether so that layer 3 addressing only reflects
the current locations within the topology.

That means you rebuild layer 4 to get node identity from some other
source than layer 3 and you propagate that change up the stack to
layer 7.


Regards,
Bill Herrin


-- 
William D. Herrin ................ herrin@dirtside.com bill@herrin.us
3005 Crane Dr. ...................... Web: <http://bill.herrin.us/>
Falls Church, VA 22042-3004

--
to unsubscribe send a message to rrg-request@psg.com with the
word 'unsubscribe' in a single line as the message text body.
archive: <http://psg.com/lists/rrg/> & ftp://psg.com/pub/lists/rrg