[ppml] Policy to help the little guys

Owen DeLong owen at delong.com
Wed Mar 19 16:06:47 EDT 2008

On Mar 20, 2008, at 2:03 AM, William Herrin wrote:

> On Wed, Mar 19, 2008 at 2:38 AM, David Williamson <dlw 
> +arin at tellme.com> wrote:
>> I'd be in favor of that.  I still think there's a lot of demand for
>> multi-homed sites that have a legitimate need for PI.  Forcing small
>> multi-homed sites to lie to get PI space seems less than ideal given
>> the constraints of IPv4 availability.  If you can get a PA /24 for
>> multi-homing, and anyone can, why can't you get PI?  It's not like  
>> it's
>> going to pollute the routing table any more than the PA /24.  The  
>> only
>> "advantage" to the PA /24 is that you get stuck with one of your
>> upstreams, whether you like it or not.  I'll also note that the other
>> RIRs already have made this shift.
> Hi David,
> The main problem with PI assignments is the systemic cost of
> announcing a BGP prefix ( http://bill.herrin.us/network/bgpcost.html
> ). If you look at the things from a different perspective, the problem
> is the lack of mechanisms by which the 27000 active AS's can recover
> the BGP cost directly from the folks announcing a prefix.
The cost of a PI assignment and a multi-homed PA /24 is identical
from a routing table perspective.  Both end up announced into
the routing table.

In fact, for "TE", Verizon and others announce (or forward the
announcement of) many PA more specifics that are not even
connected to more than one independent AS.

> The prefix length restriction at the RIRs is a somewhat upside-down
> way of assuring that few who aren't putting lots of money into the
> infrastructure can consume the expensive resource. To justify a /22
> you must document a fairly sizable IT operation, one whose nature will
> require spending at least tens of thousands of dollars a year on
> Internet transit.
Not true.  One can easily support sufficient IT infrastructure to
legitimately utilize more than 512 host addresses for just over
$100/month if one is frugal and has a relatively low traffic per
host footprint.

> Solve either variant of the BGP cost problem first (cut the cost or
> figure out how to bill for it) and there'll be no reason for ARIN to
> maintain a limit on minimum assignment size.
Since the BGP cost problem is identical in the case of
a multi-homed end-site whether they get their space
from PA or PI, I don't see how this is a legitimate argument.

Indeed, instead, it seems to me that the current restrictions
constitute a Pony for the incumbent ISPs supporting provider


More information about the ARIN-PPML mailing list