[ppml] Legacy /24s

[Owen DeLong]

> Leo,
>
> Lets put some numbers to this so that we're arguing facts rather  
> than opinions.
>
> There are 233,000 IPv4 routes and a little under 1000 IPv6 routes in
> the default-free zone (DFZ) today.
>
Agreed...
> 10%-30% of the routers in the DFZ today have a hard limit between
> 244,000 and 260,000 IPv4 routes or half that number of IPv6 routes.
> When the limit is reached within the next few months, those routers
> will experience various degrees of falling over dead. Upgrading these
> routers to accomodate 1M IPv4 or 500k IPv6 routes will cost
> $30,000-$150,000 each.
>
I'm not so sure about your estimate of the number of routers in the DFZ
that would need to be upgraded or where you came up with that number,
but, for the moment, I can go with it.
> In the more general case, these routers have to be upgraded every  
> 3-5 years.
>
Most internet-oriented technology needs to be refreshed on a 2-4 year
life-cycle, actually.
> The number of routes and ASes in the DFZ implies that there are
> somewhere between 200,000 and 300,000 routers in the DFZ.
>
Here, I think you go off the rails with a big hand-wave.  How, exactly
do you think you can correlate the routes+ASs in the DFZ to the
number of routers participating in the DFZ?
> Lets start with the low numbers. If its $30k to put 233k routes in the
> DFZ then each route in each router costs around 13 cents. Times 200k
> routers offers an aggregate worldwide cost of $26,000 to announce an
> IPv4 route. The IPv6 routes consume twice the capacity in the router,
> so that means the worldwide cost of announcing an IPv6 prefix is
> $52,000. On a 3-year upgrade cycle that's an annual attributable cost
> of $17,300 per prefix.
>
Why is an IPv6 prefix 2x an IPv4 prefix?  It's 4x the prefix size,  
but, given
all the other path attributes and such, I still don't see that  
doubling the
storage requirements.

> Lets try the high numbers. $150k to handle 500,000 IPv6 routes = 30
> cents per route. Times 300k routers = $90l. Divide by a 5 year upgrade
> cycle = an annual attributable cost of $18,000 per prefix.
>
While I can potentially accept your price per route per router, I am not
sure I accept your computation of the number of routers it is multiplied
by.  This means that your $17-$18k per prefix per anum number is
very suspect since the 200k routers is a much bigger component
of the end number than the price per route.
> While this is not a thorough cost analysis, its reasonable for
> ballparking. I can say with an acceptable degree of confidence that
> the worldwide attributable cost of announcing one IPv6 prefix is
> between $17k and $18k per year.
>
I have less confidence in the number than you do.
>
> So, with any proposal to expand the availability of IPv6 PI, the
> question that should be asked is: "Does the proposed use in the
> expansion justify asking the rest of the world to pay $17k?"
>
Since that's 17k divided amongst 30k or so organizations, you're
really asking each other org. to foot a $0.50 bill per prefix.
>
> My opinion is that in the case of a multihomed content provider, the
> answer is yes. Why should he receive poor treatment in IPv6 merely
> because his mechanism for providing content was efficient enough to
> require only a few IP addresses? If you're willing to pay the prices
> that the network operators require to be mulithomed (whatever that
> happens to be) then you should be eligible for the necessary IPv6 PI
> assignment. That's just how the Internet functions and the $17k is a
> cost of doing business.
>
What about a multihomed content consumer?  Why should they
be treated any worse than a multihomed content provider?
> In the case of single-homed folks who just want to avoid renumbering
> hassles, I think you'd need an awfully large number of computers
> before the renumbering hassle adds up to $17k/yr. Many many computers.
>
You are missing a fundamental factor here.  The pain/cost of renumbering
is only marginally proportional to the number of hosts you own.  It  
is much
more affected by the number of hosts with your IPs in their  
configuration
files that you do not control.
> As for the folks who want to weasel out of the $100/yr annual
> maintenance fee for an IPv6 PI assignment, I would ask why anyone
> should take your request seriously when you're simultaneously asking
> the rest of the world to spend $17,000/yr on your behalf.
>
Nobody really cares about the $100/yr.  Proposals to waive that
have been targeted at providing what little incentive we can towards
other behaviors for the common good.

Owen