[ppml] Draft 2 of proposal for ip assignment with sponsorship

[william at elan.net]

> But regardless of where the allocations come from, there will still be lots 
> of allocations that cannot be summarized that otherwise could be summarized.
Are you so sure summarization is such a good thing? Multihoming is the way to 
allow multiple routes to a network and to allow network to be up through 
backup provider when primary one is down. Now you summarize the routes 
and primary provider of some organization is down  (assuming primary is 
the one who has that /14 and company is using /22 out of that) because 
you summarized the routes you would not be able to reach that companies' 
servers. So you customers are not able to get to that network and your 
competitor next door are because they do not do summarization. Who is the 
winner here, your users or your competitor's users?
 
> >
> > In addition I have to point out again that both APNIC and RIPE are making
> > assignments smaller then /20 out of their blocks which seems to indicate
> > a support of this among majority of RIRs. And this also means no matter
> > how we look at it, smaller announcements would still dominate routing
> > table if the grows of internet in the countries that represent 90% of the
> > World population continues.
> 
> I am not intimately familiar with RIPEs or APNICs policies.  However, 
> according to RIPE-269 you are not correct:
> 
> IPv4 CIDR block 	Default RIPE NCC	    Smallest RIPE NCC
> 			   Allocation	    	Allocation / Assignment
> 	 62/8 			/19 			/19
> 	 80/8 	 		/20 			/20
> 	 81/8 			/20 			/20
>          82/8			/20			/20
> 	193/8 			/19 			/29
> 	194/8 			/19 			/29
> 	195/8 			/19 			/29
> 	212/8 			/19 			/19
> 	213/8 			/19 			/19
> 	217/8 			/20 			/20
How am I not correct if on the same RIPE document it says "Allocations or 
assignments smaller than the default size have been made to users 
requesting Provider Independent (PI) IPv4 address space." (did I see /29 
there above?).

And do note that by RIPE's definition and the name itself LIR (Local 
Internet Registry) does not necessarily have to be an ISP, it can be 
organization dedicated to subdelegating space to ISPs somewhat in a way 
RIR does (so they could fragment the space in routing table, but this is 
obviously discoraged).

But to be more to the point RIPE has a procedure in place on how LIR can
do micro-assignments (they call it Provider Independent Address Space):
http://www.ripe.net/training/lir/material/slides/page103.htm
http://www.ripe.net/training/lir/material/slides/page106.htm
By that procedure LIR can request PI space on behelf of the end-user and 
then RIPE assigns that space to LIR and LIR to the end-user. This procedure
seems somewhat similar to what I'm proposing in that ISP as direct contact 
with end-user first checks the request and then sends it to RIR and RIR 
assigns the space, the difference is that in my proposal end-user becomes 
client of RIR where as with RIPE's system, end-user is still a client of 
LIR but can change to different LIR.
 
> APNIC does have a policy for multi-homing though I am not familiar with it. 
> I am sure somebody else here can provide more accurate information.
APNIC person already answered confirming that they have specific policy
for micro-assignments and its generally more liberal then ARIN's current
proposals.

> > And as far as load on the routers we have Moor's law that says that they
> > would become twice as fast every two years, the grows of the routing
> > table  as seen at http://www.employees.org/~tbates/cidr.plot.html is a
> > lot smaller and everything does indicate that routers are becoming faster
> > and smarter and more capable and new technologies are also being invented
> > that help to  deal with more complex routing table.
> 
> Well, routers are getting faster and smarter, but you are assuming that 
> routing table growth and routing table computation increase linearly with 
> respect to one another.  I do not know enough about route computation 
> algorithms to say one way or another.
More research is indeed needed, but so far nothing has suggested that 
non-linear situation exists if anything things in computing world with 
large databases show a different trend. But having no data on the issues 
does not allow use it either way in this debate.

> But can multihomed joe end user afford the latest and greatest 6 and 7 
> figure router?  Many people are running low-end or old routers that are 
> being pushed to the limit.  What about them?
Upgrade - equipment is cheap these days. Somebody not upgrading and still 
being in the "stone age" should not be keeping us in the same age! New 
technologies are being created and if anything ARIN should encorage
technological progress and not the other way around.

And just on this point, the proposal would not increase size of routing 
table at least not to those who do not do summarization and for others 
not right away in any considerable amount giving them enough time (i.e. 
probably 2-5 years) to upgrade and take advantage of new router hardware 
that will be needed anyway if they want to speak ipv6 in the future.

> Just so we're all on the same page, I am all for doing what is best for the 
> entire Internet.  I realize I have mainly been arguing routing table issues 
> against reducing the minimum size, but that is mainly because of the fact 
> that there are very few people arguing those points, and I feel it is 
> important for both sides of an argument to be presented.
So far I have not seen conclusive evidence indicating that multihoming 
organizations should continue to use their upstream providers's ip blocks.
This in itself goes against all purpose of multihoming!

> Alec
> 
> --
> Alec H. Peterson -- ahp at hilander.com
> Chief Technology Officer
> Catbird Networks, http://www.catbird.com