[ppml] Revising Centrally Assigned ULA draft

mack mack at exchange.alphared.com
Tue Jun 19 23:09:30 EDT 2007


A number of people had a number of scathing remarks regarding my last post:

1) If people want to use IPv6 NAT let them.
Some are going to because there are going to be people marketing equipment for it.
I don't honestly think anyone is going to be using PAT with IPv6.
If they are then they are seriously brain damaged and should not have access to a computer.

2) The number of routes in the table is going to be an issue.
Cisco has a route processor that will handle 1M routes IPv4 routes and 512K IPv6 routes.
It will be at least 2 years before the next iteration is ready I am fairly certain.

The price tag of this device is rather high and requires every
blade to be upgraded with a new DFC board as well.
Multiply that buy the number of routers required to carry full routes.

We aren't planning on upgrading this year.
We are stuck with the current 512K IPv4 and 256K IPv6 routes limit.
We currently see about 220K global prefixes in IPv4 and 845 in IPv6.
The total number of prefixes is higher, but those are the ones seen from a majority of providers.

3) The BGP loads on edge routers are quite high.
It takes several minutes for a full BGP reload of one session.
Bad things can happen if multiple BGP sessions reset at once.
I have seen that happen.  It isn't pretty.  This isn't a future problem.
This is happening now.

4) /48s are going to get filtered if the number of routes gets too high.
That is a fact of life.  If the router won't handle it, then it won't get routed.

5) Criteria for /48s to people who have a business need (as opposed to routing need) is a different policy issue.
IP addresses based on business need would make a good policy proposal should someone choose to write one.

Basics of a business need proposal for IPv6 only:

a) Entity must be capable of routing the IP block assigned.
(This includes technical capability as well as owning equipment and
having an appropriate contract with an upstream)
b) Entity must be a business or non-profit organization as defined
by some set of criteria (Tax ID # perhaps).
c) Entity must show that renumbering its IP space on changing transit
providers would be a significant financial burden on the entity or its customers.
(some minimum computer count and other statistics)
d) Entity must be a unique site from any other entity granted IP space.
(no double dipping or having a single web server in someone's data center).

The arguments against ULA-C seem to be:

1) It is a waste of space (how big are those nanobots again?).
2) People will use it for NAT (which is their own problem).
3) People have a business case for IP addresses (which is a different issue).
4) Everyone should be able to get a /48.

Why do the same people who argue ULA-C is a waste of space think everyone should have a /48?

Summary:

As someone who has to deal with this on a daily basis, I would prefer to see aggregated routes.
ULA and ULA-C is the appropriate place for giving out unique /48s to anyone that wants them
and can't/won't justify them.

If someone thinks a business need proposal is worth fleshing out please do so.

LR Mack McBride
Network Administrator
Alpha Red, Inc.




More information about the ARIN-PPML mailing list