[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Site local

Kurt (or should I say Kurtis?) raised the question how many multihomers

Everyone I know calls me Kurtis..:)


there are now in IPv4. There are three answers:

1. We don't know
2. Not all that many
3. It's not relavant, it's the long-term growth that's the problem
Well, I think you misunderstood the point I was trying to make.

We know that the growth of the number of routes in the routingtable have slowed down. AFAIK we have no clear understanding of why, but there are in principle two options. The burst of the dot.com bubble or better operating practices of ISPs.

What is a problem is that we don't know how from where multihomed prefixes are coming. Basically it could be

- Having been singlehomed announced PI space going mulithomed
- Break out blocks of PA space coming on-line as multihomed
- Newly allocated PI space becoming multihomed
- Newly allocated PA space being multihomed.

All of these are possible and more or less still imply enterprises. So, most of these are possible today but we don't know the growth rate, and actually that is pretty hard to find out.

Now, I think we can agree that there is a second group of possible multihomers in home users. What we don't about them is how many are actually willing to pay for multiple connections? What will the providers charge for a multihomed DSL solution? How will this affect growth? Remember that nomatter what solution we come up with it will generate extra costs at the providers that will be passed on to the customers. Maybe this is the solution to the multihoming problem....

One reason the current number of multihomers is relatively low is
probably because it is relatively hard to do. Another is the
I disagree. All you need is two multiple connections and the RIRs will give you a AS number.

availability of alternatives. Both of this has the potential change. As
more and more people start to multihome it will become a matter of
routine, this can become a snowball-effect. In IPv4, people have
relatively small address blocks and use NATs. In IPv6, the address
blocks are huge and not as many people will use NATs, so even though
Now they will use globally unique site-local addresses...:) (Sorry, couldn't resist..)


But now there is an interesting development in the IPv6 working group:
they reached consensus it is a good idea to look at globally unique,
non-routable (although this part was immediately challenged) address
space to replace/complement current site local scoped addresses.
Well what I thought we voted for was to rip out the scoped address part completely as alternative 0, and that we where going to work on alternative 1 and two. The from nowhere we get the second address space...sigh. Anyway, that is off-topic for this group.


If large enterprises can use this type of address space for all their
internal stuff, renumbering becomes much easier as there are no security
issues: globally routable addresses from an ISP are never trusted, use
site local for internal stuff = no need to change filters when
renumbering. The main renumbering issue that remains is that of the DNS
interaction, maybe along with pushing a new /48 down the internal
network. These seem relatively solvable.
(Notice that the following is irony and I STRONGLY recommend to take this seriously)

With SLs and NATv6 we have solved the multihoming problem as each enterprise then only need a number of PA addresses. We might have broken the application layer but that is not our problem :)

I think that someone will need to start co-ordinating this IPv6 stuff between the working groups....


Does this mean we should annex the globally unique site local effort and
work on that in this wg?

My vote is a clear NO - for those of you who hadn't guess....

- kurtis -