[ppml] Fw: IRS goes IPv6!
On 02/25/06 at 9:58am -0500, Joe Provo <ppml at rsuc.gweep.net> wrote:
> On Wed, Feb 22, 2006 at 09:37:48PM -0500, Scott Leibrand wrote:
> > While tier1 status is indeed a lot about ego, it is quite
> > useful in a lot of technical discussions. For example, the definition of
> > a tier1 NSP is someone who peers (interconnects) with all other tier1's.
> > (I'll leave settlement out of my definition, as it's not useful in
> > technical discussions, just political and ego ones.) If you'd prefer a
> > less overloaded term, perhaps "transit-free NSP" would work.
> Those terms both fail.
> ISPs are business entities which receive registry allocations. These
> business entities operate many ASNs with differing policies,
> unfortunately not always delineated by AS boundary. While an ARIN-sphere
> business entity may operate a transit-free network or portion of their
> network in the ARIN sphere, it may operate a transit-using network or
> portion of their network outside of said sphere... and vice-versa.
Ok. I'll agree that "tier1 NSP" and "transit-free NSP" don't fully
describe the world we live in. But pretty much any label break down for
some cases: that doesn't mean we stop using labels when talking in
> It seems to me that trying to tie an allocation policy to a routing
> policy (which will change over time) is merely adding:
> - admin overhead to the registry
> - more stuff which the registry will have as rules and yet have no
> method of enforcement (revocation anyone? great stability there)
> - more admin overhead (cost) for the receiving ISPs that play along
> - zero incentive for the receiving ISPs to actually play along.
> We've seen that work *so* well before.
Ok. I would agree with everything you say, but I'm not sure why you're
saying it. I'm certainly not proposing to tie allocation policy to
routing policy, nor have I seen anyone else suggest that recently.
I'm simply proposing that ARIN allocate PI space in a systematic manner
rather than on a random (chronological) basis. Any system based even
loosely on topology or geography will allow well-connected NSP/ISPs to
aggregate within their own network if/when they feel the need.
In such a system, RIRs will not have to have rules or enforce anything.
Overhead can be minimal: as simple as providing the address provided by
the applicant and the netblock size needed to a simple algorithm that
returns the best free netblock to allocate. There need not be any
overhead whatsoever for receiving ISPs. No one need "play along", so no
one would need to encourage anyone to do so. ISP/NSPs need not change
their routing policies unless they feel it would save them money to do so
(in delaying the need for CapEx spent on router upgrades).
> In the Internet-wide scope, why is being transit-free --or more
> accurately, backup-free and disaster-vulnerable-- important? Those
> networks are *just* as reliant on third parties for carrying their
> traffic as transit-buying networks. Multiples of those networks
> have had serious financial problems resulting in previous and current
> acquisition by networks built along different policies that have
> been successful in business. I guess I'm wondering why we'd want to
> encourage f[l]ailing models.
There's no need to treat transit-free peering-only networks differently
from anyone else. However, since such networks provide transit for
everyone else, we do need to ensure that we aren't asking them to do
anything that would prevent them from maintaining global reachability.
> The set of networks carrying a DFZ is larger. The set of networks
> ARIN represents is even larger. Pushing to centralize forwarding
> decisions affecting all of them in a manner counter to their economic
> and stability concerns is poorly thought-out, not to mention casts
> a shadow of a bell-shaped head.
Heh. I've never been accused of talking like a bell-head before. Ma Bell
was broken up before I was born, and by the time I got to high school, the
Internet boom/bubble was in full swing.
I'm not pushing to centralize everyone's forwarding decisions. I fully
appreciate the benefits of being able to interconnect and peer with anyone
you want, and getting all of their routes to ensure you use the most
direct path. Aggregation by transit-providing networks wouldn't change
that in the slightest.
However, there's a reason ISPs buy transit: they can't afford to build out
the global network required to interconnect with everyone they might wish
to exchange traffic with, so they pay someone to deliver their traffic
where it needs to go. Keeping that in mind, I don't care whether my
transit providers use an aggregate or deaggregates to take my traffic to