[arin-ppml] IPv6 /32 minimum for extra-small ISP

[Steve Bertrand]

On 2010.04.14 18:54, Paul G. Timmins wrote:
> 
>> Thanks. Real data. Seems like a useful contribution to the discussion.
>> Not sure why the questions leading to this point generated such
>> hostility, nor why it had to be preambled with "snarky" stuff.
>>
>> So the "every router" statement has now dwindled to "most large ISP
>> routers and about half of small ISP routers, which together compose
>> about 10% of the world's routers." (Could be a misreading of the data
>> because we don't know whether Randy's #tested is representative of the
>> proportion in the total population.)
> 
> Another thing to consider is that just because someone has a default route doesn't mean they don't carry full tables. A recent acquisition by our company had two routers with full tables to different ISPs, but there was still a floating static route configured.
> 
> I think this is probably pretty common - admins who aren't BGP professionals not wanting to trust their BGP configurations entirely in the event they do something wrong. But there are definitely costs to carry the routes, regardless if they use a floating static or not.

I can't differentiate between NANOG and ARIN PPML anymore, so what the
heck :)

What I haven't seen noted yet, is that there is also a cost to
_filtering_ a route that you don't want.

If you aren't AS-HUGE who can afford to accept only very large prefixes
until they find out whether it makes practical sense to do otherwise, it
costs the same or even more to keep a route *out* of my routing table.

Whether it's accepting a prefix in BGP, or having a default route where
one has to create ACLs to block traffic to a destination, there is a cost.

I'd like to think that even a small enterprise would maintain a list of
destination prefixes that they don't want their users going to...when
they do, where do they apply the rules?

Do they lower costs by maintaining the rules only at the outside
perimeter, and then deal with bandwidth issues when they have to focus
on excess traffic transiting the network until it hits the ACL? Or do
they spend on documenting the network clearly, applying sound practices
on each device on the network?

...I'm just curious. I see that there is a cost to each new prefix,
whether BGP is involved or not.

Steve