Verio Decides what parts of the internet to drop

Austin writes:

George writes:

I am not sure whether the danger in opening up the B space for
/17 blocks is particularly bad, but lacking a single consistent
policy body with sufficient clue about both the Tier-1 backbone
issues and the address allocation issues, it's hard to fault
any given ISP for insisting on /16s in B space.

Sounds good, but what exactly does that mean? Does any end network
capable of justifying a /24 then get a routable chunk, thus blowing up the
tables? What if you could do it based upon traffic generation? That would be
difficult to verify, and the definition for 'large' amounts of traffic is
ever changing.
So, if we say that a /20 is a sufficiently large amount of space to
get a routable chunk, then they would be able to get it from ARIN anyway,
and we're back to square one.
In the far term as space becomes scarce we will need to find a solution
to wasted B space, but that is several years out. Perhaps by that time routers
will have so much memory and CPU as to make an extra ~4 million possible routes
negligible.

The danger of /17 blocks in B space is limited to 64*256 more routes
(16 k more, maximum). All at once that would be bad, but over time
it would be reasonable. I would personally, were I setting route
policy at a Tier 1, allow a /17 in B space, but there's no reason
to try and force anyone else to accept that. As others don't
right now and aren't inclined to, I would dissuade anyone from
trying it as it's either going to be a royal pain or impossible
to get routing for.

Again, this is where not having a single policy body is killing us;
some people get oddball blocks, somehow or another, and are then
screwed on routability. I am most certainly not going to suggest
radical surgery to the current way of doing it; right now, small
places deserving of multihoming have to work at it, and you have
to be clued enough to not step in a few holes like trying to
portably subdivide a B. I am not sure that there's any obvious
fix in the nearterm for those problems, which are avoidable.
As long as they're avoidable I think the thing to do is to
leave well enough alone.

-george william herbert
gherbert@crl.com
Disclaimer: I speak for myself only, not my ISP, Cats, wife, or car.

In the far term as space becomes scarce we will need to find a solution
to wasted B space, but that is several years out. Perhaps by that time routers
will have so much memory and CPU as to make an extra ~4 million possible routes
negligible.

The danger of /17 blocks in B space is limited to 64*256 more routes
(16 k more, maximum).

  Yes, you could arbitrarily say /17 is a fair border, and then people
would complain about their /18s being unreachable. The 4 million number
reflects 64 * 2^16 theoretical /24 routes - 64 * 256 current theoretical /16
routes = 4177920 routes. I haven't heard (yet) of people complaining about not
being able to get /25 to /32 routes globally routable.
  Perhaps a somewhat less arbitrary limit corresponding to the smallest
allocation made by ARIN would be in order. That would currently be 2^(20 - 16)
* 64 * 256 - 64 * 256 = 245760 extra routes. Still a pretty highg number, but
I imagine it would take several years to break up the existing Bs.

  Austin

Yes, you could arbitrarily say /17 is a fair border, and then people
would complain about their /18s being unreachable.

using the rirs' allocation boundaries is logical, protects against some
of the worst disasters we have had, and is a clearly visible detent on
the dial. beyond that is a slippery slope with no rationale i can see
for any stopping point until one hits /24, which we know invites global
disasters.

randy

And what I'd really like to know: how many millions and billions were spent
by domestic telcos to accomodate and ultimately deflect anti-trust action
heading their way regaring local, 800 (and soon: cellular, at last!)
number portability ? (lets call it xNP)

I mean: there must be an order of magnitude of increased HD space,
RAM and SS7 network bandwidth in use right now due to xNP.
Which means that the telcos probably asked their vendors to provide
such capabilities for their switches - and got what they asked for!

Lets face it: if the US PSTN can accomodate tens of millions of essentially
freely-routed (well, the stubs of the SS7 network are certainly very static,
heh) phone numbers, it must be possible to scale the Internet beyond such
a small pisser: a 1/4 million routes in the BGP table.

Given that more and more end-user organizations realize that it's
impossible to do proper large-scale business on the Internet without
"cheating" allocation policies in gross and wasteful ways in order to
create proper load-balancable (uh, I am sticking my head out here)
multi-homed networks, a change in attitude amoung us implementors and
R&D folks is in urgent need: are we constrained merely by our small minds,
equipment limitations and current software implementations and
protocols, or have we indeed hit a fundamental brickwall with BGP-4,
as some scary early findings of CAIDA seem to suggest ?

As network operators, I think we should prepare for the equivalent
of the US running out of 10-digit phone numbers, a situation that
might make Y2K look like a footnote in global telco history:

- IPv6 is not the answer to our routability problems, but it will vastly
   accelerate the reachability problems we already have. Provider-based
   prefixing will be a breaking dike once it becomes obvious to people
   that geographical or organizational hierarchies cannot be dictated
   over business needs.
- organizations must be relieved from wasteful and expensive renumbering
   processes as much as possible, especially since organizational growth
   will essentially be infinite, either in numbers of organizations,
   or hosts connected per organization.
- there will be exactly one road to Rome: one organization, one route per
   logical location. Read my lips.

There can be no denial that this is where things are going. You may not
like it, but this is where its headed right now, with all the ugly
side effects of IP space waste and cheating on allocations just to
overcome some basic operational problems.

Lets start preparing for this, as we will do this not because it's easy
and apparent, but because hard business-needs are going to drive us this
way in at most a year or two, with overwhelming benefits to endusers of IP
space outweighing all efforts to overcome the current limitations.

All ends and odds on this are open, as far as technology, implementation and
settlement models (if any) is concerned, and I'd welcome someone experienced
with setting up an IETF WG stepping forward. Title for such a WG ?
"Internet Routing and Address Space Use of the Future". If such WG fitting
such an agenda already exists, please kindly point this out to me.

Thank you.

One quick difference - the PSTN only has to make a single routing decision
per call. Internet routers usually decide on a per-packet basis.

- IPv6 is not the answer to our routability problems, but it will vastly
   accelerate the reachability problems we already have. Provider-based
   prefixing will be a breaking dike once it becomes obvious to people
   that geographical or organizational hierarchies cannot be dictated
   over business needs.

Agreed. Sounds good in theory, but I wouldn't bet on most organizations
being happy with keeping these hierarchies intact.

- Steve

The phone system doesn't require anything close to millions of routes for
LNP. Instead, at the time of call setup, there is a lookup that performs
the translation between the portable number (which is the logical address)
and the physical address (which to date is still mostly statically
routed using a well-defined hierarchy based upon physical location). The
LNP translation is quite akin to that of a DNS lookup, which is again
done at the beginning of an IP "conversation".

Prabhu

Kai Schlichting wrote:

The phone system doesn't require anything close to millions of routes for
LNP. Instead, at the time of call setup, there is a lookup that performs
the translation between the portable number (which is the logical address)
and the physical address (which to date is still mostly statically
routed using a well-defined hierarchy based upon physical location).

and here is where the anology breaks down. a second or two of call setup
may be acceptable for establishing a phone call. it would be a disaster
on a per-packet basis.

ip is a connectionless protocol. before hitting the reply key, think about
that.

randy