- Date: Tue, 27 May 2003 10:17:21 +1000
>From: Gert Doering 
>Sent: 23 May 2003 07:07
>To: leo vegoda
>Subject: Re: Continuation...
>As far as I understood Kurtis, a major reason for going PI is
>that a new LIR might not be able to get their allocation, and
>doesn't want to use someone elses PA space ("inofficial
>sub-allocations"), thus they go for PI.
>The idea behind requiring a minimum usage for the initial PA
>allocation was (IIRC) to slow down the growth of "new LIRs"
>somewhat, and be conservative in PA address space giveaway.
Since I was part of the original discussions around the PA allocation
criteria, I thought I should add a few thoughts.
One reason PA allocation criteria were introduced was to make policy
more transparent and clear. Instead of having the RIPE NCC hostmasters
trying to convince people who only really needed a few addresses to not
get a /20 allocation, it was agreed that clear, documented criteria were
Another argument that was raised, was that it didn't make sense that you
needed to justify each and every assignment, while anyone could get an
allocation without showing *any* need for that address block.
However, at the time, we did also conclude that we were only solving one
part of the problem and that we still needed to discuss PI assignment
policy as it is all interrelated.
>Now this seems to backfire, and hurting *more* in the address
>space fragmentation ("lots of small PI stuff", instead of "one
>/22 PA") and routing area.
>What about the following way to tackle it:
> - we change the PA policy back to "no initial usage to be documented"
> [- eventually reducing initial allocation size to a /21]
Hm. I'm not sure if I follow you Gert. In what way is it better to give
an organisation a PA allocation rather than a PI assignment? They both
add the same amount of routes to the routing table. The only difference
is that people might get more than they need if they get an allocation.
If someone wants about 500 "portable" addresses, only for their own
network, that they can multihome with, isn't it better that they get a
/23 PI assignment than a /20 PA alloction...? After all, PI is *in
itself* not bad for the routing table. It's the concept of everyone
wanting to be independent and multihome (as well as people chopping up
their PA allocations) that adds to the growing global routing table.
> - the PI form gets a yes/no field "routability needed"
The risk with adding this to a request form is that you are implying
that this routability can be guaranteed, which the RIRs cannot.
> - the PI form gets a "if you need routability, please state
> why you can't use PA space" text area (if people want non-routeable
> space, it's for internal stuff that has nothing to do with
> so PA would be the wrong answer)
Yes, the requestors should of course make an informed decision. I think
the RIPE NCC hostmasters do inform PI requestors of the alternatives as
well as the disadvantages of PI (last time I looked, but Leo might
correct me here.:)
I understand that what you're trying to solve with your suggestion is
routability. But there's a risk with the RIRs saying: "Ok, you need
routabilitiy? Then we'll give you these addresses and you'll be fine!"
I think the original reasoning around PI was "The RIRs has no power over
routing. If people want PI, they can get it - as long as they are aware
of what it does to the routing table and that they risk getting filtered
out." But I also understand the argument Kurtis had about the policy not
working if the RIRs are giving out "unusable" address space.
A suggestion that has been made is to set a minimum PI assignment size.
On one hand it does go against the principle that all addresses should
be justified. On the other hand, it addresses the issue of routability,
increasing its chance of being routed without the RIRs guaranteeing it.
(Although the occurance of PI requests smaller than /24 is quite rare.)
But even this isn't straight forward. What minimum size do you set? /24?
/23? - And then what if providers start filtering on larger prefixes?
The big question is really what we're trying to solve here. Finding ways
of reducing the routing table? Or finding ways of increasing small
organisations chances of being routed?
I'm afraid I'm not providing any answers here. I'll leave that up to you
wise people to do... ;-)
>So let the discussion start...
> -- NetMaster
>Total number of prefixes smaller than registry allocations:
>SpaceNet AG Mail: netmaster@localhost
>Joseph-Dollinger-Bogen 14 Tel : +49-89-32356-0
>80807 Muenchen Fax : +49-89-32356-299