[ppml] NANOG IPv4 Exhaustion BoF

John Curran jcurran at istaff.org
Fri Mar 7 02:59:53 EST 2008


At 5:39 PM +1100 3/7/08, Geoff Huston wrote:
>Yes, the situation the this industry has driven itself into is messy, but
>I'm pretty sure that global collapse is not exactly on the agenda.
>...
>What I would claim is that this transition is just another risk factor
>in an industry that has its fair share already and doubtless will
>continue to have its fair share in the future.

I'd agree with you if there were anything resembling a useful
feedback mechanism on the global routing table.  As it is, the
only knob available is route filtering, and while that works fine
as a tool against needless deaggregation, it really doesn't work
well filtering lots of new unique routes from your peers all of
which are necessary to cover their new customers...

The costs of injecting new routes are borne by third parties (i.e.
all other DFZ participants).  Without direct predictable routing
costs, there is no counter-pressure to fragmentation in the
presence of a high demand marketplace.

If ISP's continue to attempt to globally connect new customers
via IPv4 after depletion, one had best hope they do it by finding
and obtaining very large underutilized blocks; the usage model
would closely resemble today's RIR-provided blocks, and the
resulting routing demand when announced would at least be
comparable to today's (manageable) situation.

If ISP's can't obtain large blocks post depletion and we instead
see customers having to BYOB ("Bring Your Own Block") to get
IPv4 connectivity, then new routes will be introduced at a much
greater rate, and with very real uncertainty regarding whether
these routes will be filtered outside of their originating ISP. 
Despite such filtering, ISP's are motivated to keep connecting
customers and refer connectivity issues to the filtering peer,
since such filtering is "voluntarily".  Some organizations would
consider connections to the majority of the DFZ ISP backbones
so that their public servers can have at least one address on the
same "Carrier Internet" backbone as any new customer site.

That end state isn't global collapse, but it also isn't global connectivity.

/John

Disclaimer:  my opinions only; discard or forward as desired.







More information about the ARIN-PPML mailing list