[ppml] Policy Proposal 2005-1: Provider-independent IPv6

[Stephen Sprunk]

Thus spake "Jason Schiller (schiller at uu.net)" <jason.schiller at mci.com>
> I am against this policy.
>
> It seems that people really want multi-homing badly to make IPv6 work.
>
> Heidi Hinden's first law: When you want it bad, you get it bad, and most
> people want it in the worst way.

We'd prefer a better solution, really, but the IETF has failed to provide us 
one.  Since multihoming is a fundamental business requirement for 
enterprises and PI is the only way to make multihoming work today, either 
enterprises get PIv6 or v6 doesn't happen.

Please keep in mind that we don't like this situation any more than you do.

> What concerns me are three things:
>
> 1. Enterprise customer who want PI addresses or useful multi-homing, and
> don't care about the problems it creates for the large ISPs that carry
> full routes.  (That's their problem.)
>
> In reality it is everyone's problem
> if they want to transit one of these ISPs, or use best path routing
> (carry full routes and not just a default to a transit provider).

OTOH, one could say that large ISPs don't care about the problems that PA 
space causes enterprises.  2005-1 is an attempt to find a compromise that's 
acceptable to both sides.

> Lets not forget that router vendors are behind the curve on port speeds
> too.  Are these vendors more likely to solve the routing table problem
> that affects only the largest ISPs or focus on port speed problems that
> affect many large enterprise customers?

Vendors, like any other business, will solve the problems that result in the 
highest profits.  It's not surprising, therefore, that the vendor-dominated 
IETF has failed to produce a new IDR paradigm that makes routing table 
growth (and thus constant upgrades) unnecessary.

> 2. The concern people are being short sited and since there are only 1,000
> routes in the IPv6 Internet table that this will not be a problem any time
> soon.

That's hardly what we're saying.  What we're saying is that the floodgates 
are wide, wide open for PIv4 yet there's a paltry ~9650 non-transit ASes in 
ARIN-land and multihoming is growing slower than Moore's Law.  There is no 
rational reason to expect that the number of ASes will increase 
significantly solely due to 2005-1 since we're reusing the existing bar; if 
there is a rush for ASNs, the v4 table will grow more rapidly than the v6 
table due to the inherent prefix-per-ASN differences.  PIv6 will be the 
least of ISPs' worries if your fears turn out to be correct.

Now, if we had set the bar for PIv6 lower than for PIv4, I'd agree with you 
that there's a serious danger of an overwhelming land rush, but that's not 
the case.  The bar is the same, and that was deliberate.

> 3. The concern that we haven't done enough research to know if the vendors
> will be able to stay far enough ahead of the route table growth to not
> have a problem.  It is not enough for vendors to build the routers big
> enough in time.  If it takes 3 years to fully replace a network, and the
> router vendors are only two years ahead of the curve, then I only get 2/3
> through my upgrades before having to start a new set of upgrades.  Never
> mind being able to depreciate the cost of the router over 5 years.
>
> We have to understand what it means to make a long term commitment to
> deaggregation.  I don't hear the six largest ISPs standing up and saying
> we did some studies of what the routing table will look like in five to
> ten years, and have talked to our vendors and we don't think it will be a
> problem.

OTOH, I don't see any ISPs standing up and saying they did some studies of 
what the routing table will look like in five to ten years, and have talked 
to our vendors and we _do_ think it will be a problem. [sic]

> The point Aaron was trying to make was in reference to my
> projections.  For example I want to buy new routers today.  It takes 2
> years to certify and fully deploy the router throughout the network.  I
> want the router to live in the network for 5 years to depreciate the
> value.  That means if by 2011 there is wide spread adoption of IPv6 the
> router will need to support 1.3M routes.  This example does not take into
> consideration L3VPN routes, or routes from converging multiple networks
> onto a single chassis.

I'd like to hear why you think you'll need to support 1.3M IPv6 routes in 
five years.  Particularly, I'd like to hear what you think is wrong with the 
pro-PIv6 camp's assertions that we should see 1-2 routes per ASN and that 
the growth in ASNs will be at or near historical levels.

If we're wrong, we'd really like to know how and why so that we can find a 
better compromise.

S

Stephen Sprunk        "Stupid people surround themselves with smart
CCIE #3723           people.  Smart people surround themselves with
K5SSS         smart people who disagree with them."  --Aaron Sorkin