Hello,

Dear diary, on Mon, Nov 04, 2002 at 10:43:32PM CET, I got a letter,
where Bob Hinden <[EMAIL PROTECTED]> told me, that...
> Even though there is no consensus on how site-local addresses should be 
> used, I think there is enough people who want to use site-local that it is 
> reasonable that the w.g. should continue trying to flesh out site-local 
> usage as well as pitfalls of usage.
..snip..
> I believe this approach should help provide the larger community (e.g., 
> vendors, ISP's, enterprise operators, etc.) the information they need to 
> make an informed decision on the usage of site-locals.

  I would like to present here one possible scenario of site-local usage, which
I believe wasn't mentioned here yet. It's not a real-world usage example yet,
but it may not be so far from that point.

  First, let me introduce you to the environment briefly. We're trying to
design a possible IPv6 network addressing architecture for a non-commercial
metropolitan (Prague, CZ) network based on wireless connections between nodes.
One of the driving ideas was to provide a connection to that network for anyone
without any charge, only for the price of the equipment (given that you'll be
able to come along with the admin of the node you want to connect to). Also,
connectivity to the internet should be provided by the network in some way -
but the basic principle here is to give anyone a free choice, which provider he
will use. Thus, various members of the network could be connected to the
internet by various providers (over the network itself, as the providers are
represented by own nodes in the network). Also, the network's internal topology
is totally independent on points' choice of provider. Note that this is a
running setup already - it's around 37 to 48 working nodes now (the current
status is at http://czfree.net/monitor/), and one to three internet providers
having own nodes in the network. It's running on IPv4 now, though.

  Now, the internal addressing is done just by giving out addresses from the
10.0.0.0/8 range, and NATing it on the provider nodes. For IPv6, you could:

  * Request own IPv6 space from a LIR - but that won't work out. The addresses
which matter to the global internet are those which are being used by the
providers. And none of the providers is going to advertise the whole network,
since it would donate bandwidth even to people who aren't its clients.

  * Just use directly the addresses your provider will give you - but that
won't work out as well. It would cause total mess in the topology, since the
nodes inside of the network wouldn't be numbered consistently. Also, a lot of
the nodes are probably not going to have any internet connectivity at all, thus
they would have no address.

  * Use some global private address block; then assign the internet connected
nodes also provider addresses (and let the underlying routing protocol handle
the routes propagation etc). However, we are not aware of any such a prefix
being available these days.

  * Use site-local addressing; then assign the internet connected nodes also
provider addresses (and let the underlying routing protocol handle the routes
propagation etc). This comes out as a natural choice, as it probably fully
fulfills our requirements and it is relatively elegant solution.

  We think that this use of site-local addressing is fully legitimate here.
What do you think? Or what alternative solutions would you suggest to this
problem?

  Kind regards,

-- 
 
                                Petr "Pasky" Baudis
.
weapon, n.:
        An index of the lack of development of a culture.
.
Public PGP key && geekcode && homepage: http://pasky.ji.cz/~pasky/
--------------------------------------------------------------------
IETF IPng Working Group Mailing List
IPng Home Page:                      http://playground.sun.com/ipng
FTP archive:                      ftp://playground.sun.com/pub/ipng
Direct all administrative requests to [EMAIL PROTECTED]
--------------------------------------------------------------------

Reply via email to