Thoughts on Renumbering? (Re: Criteria iPA-A) (both IPv4 and IPv6)

[Aleksi Suhonen <]

Hello,

I'm initiating this side tracking conversation on the EOF list
since I feel it would be illplaced on the lir-wg list.

Abstract: I present some of my own experiences with renumbering
and urge readers to share their's. I wonder if IPv6 will really
make renumbering any easier.


Quote from a thread on lir-wg:
} (4) Should organizations which are using a relatively small amount of
} address space be required to renumber in order to recieve a PA allocation
} from RIPE?

I think every growing organization should be required to renumber
at most once after they have received a routable block of their own.

I've been involved in efforts to renumber three or four /19's and
I feel that the common conceptions on how hard or easy renumbering is
don't reflect reality.


* To the casual passer-by the biggest problem in renumbering will
undoubtedly be the changing of network configuration parameters
of hundreds or thousands of hosts and since even modern operating
environments seem to be beginning to be able to handle this without
the need to reboot it seems to be a straight forward task, right?

The IPng working groups have put considerable effort into trying
to make this aspect of the problem go away completely. Using DHCP
will make the problem go away in IPv4 too, but you can't use
it everywhere.

* Renumbering the administrative overhead per IP (inet-num
objects at RIPE and respective accounting objects in the ISP's
own database) should be even easier. You could theoretically
just move the networks from one block to another while keeping
their relative order intact. Sensible ISPs will wish to reorder
(for routing efficiency) and resize (for space efficiency) the
allocations though since renumbering is the best opportunity for
such often neglected work.

* The first problem in renumbering that will require thinking
comes with routing equipment. Renumbering router loopback
interfaces and inter-router links will wreak havoc on the IGP
and if the network is geographically widely distributed, doesn't
contain redundancy (networks eligible for renumbering seldom do)
and the number of engineers to do the work is limited, the plan
can get ugly.

Renumbering "hosted customers" will require synchronization
between the ISP's router guys and the customers' sysadmins.
In some cases it will be possible to use secondary interface
addresses on the router so that both the old and the new network
can coexist for the short period of time that the renumbering
should take. However, I've found that the amount of time
customers need seems to be an impossible equation:

If you try to be strict and give them a relatively short time,
they will complain that their sysadmin was on an industry symposium
in Singapore and their netadmin was on a holiday cruise on the QE2.

If on the other hand you give them too much time, they will
forget about it and complain that their connection has stopped
working when you finally disconnect the old address space.

The amount of time for one customer to forget about renumbering can
be shorter than the holiday of the resposible customer's admin's holiday.

* Next you will be noticing problems on higher protocol layers.
You'll find access-lists that need updating and certain services
will need their configurations checked. Reports of problems of
this type will probably trickle in weeks after the renumbering even
with the best planning.


I consider all of the above non-issues.

* The real issue in renumbering is in my opinion the domain
name service. No, not the trivial zone data. That gets handled
as "the administrative overhead per IP" above and the IPng WGs
have made nice innovations regarding that too. What I'm talking
about here are the name servers within the address block to be
renumbered.

Changing the name servers for the reverse mapping is thankfully
trivial since you need to only work with the organization(s)
that handed you the addresses.

Changing the name servers for the forward mapping is what I see
as the problem here. Just a few years ago changing the name
servers for a SLD of a gTLD was a headache and while that may
have become easier now, the different ccTLDs will continue to
have different types of procedures and response times for
requests of this sort. Even a small ISP can easily have hosted
domains in half a dozen different TLDs and going through them
all and getting all the relevant name servers changed in them
in a timely fashion can be a formidable task.

I don't think IPv6 can change this.


As I said in the beginning of this message, I am interested
in hearing your experiences and views on this matter. I am
relatively sure that there have been earlier discussions on
this topic but as the technology evolves and matures some
of these problems disappear or get replaced by other problems
so it won't hurt to talk about them again every now and then.

--
	Aleksi Suhonen / Axu TM Oy