ipv6 addressing - non-routable equivalents?

Jonathan Guthrie jguthrie@brokersys.com
Sun, 8 Oct 2000 09:44:08 -0500 (CDT)


On Fri, 6 Oct 2000, [ISO-8859-1] Peter Håkanson wrote:

> There is no such thing as GLOBLLY UNIQUE IEEE adressess available everywhere!
 
> In a restricted sence one may rely on "unique" MAC-adresses. That is
> if certain technologies are used on link-level, restricted to a 
> subset of it's capabilities, and noone ever makes any administrative
> mistakes (or plain cheats).

...and you're willing to diagnose odd errors, sometimes.  (I've actually
met people who have received two Ethernet cards with the same MAC
address.)
 
> But in the long run we cannot build network infrastructure on 
> the assumption that "linklayer technology never changes".

> Just think of ATM. It uses no IEEE adresses. One have to create
> one, (and how do i create a unique ?) same goes for ppp links,
> Frame-delay links, mobile UTMS phones etc.

> This is one of the design "features" of ipv6 that prevents deplayment.

I disagree.  There is absolutely nothing preventing anyone who finds the
automatic mechanism to be too confusing from using addresses that don't
have anything to do with IEEE anything.  They still work just fine, and
anyone can use them, if they choose.  What's holding back IPv6 is this:  
No commonly-available routing equipment, no commonly-available remote
access equipment, and no commonly-available end user software.

I own an ISP.  A little "one-lunged" ISP.  My business is just the sort of
enterprise that can derive the most benefit from the widespread deployment
of IPv6.

Now, I've got my network on the 6bone, and I've even set up a couple of
tunnels to provide 6bone access to the places I most commonly access the
Internet from, but that doesn't help my customers because none of the
equipment I use to deliver Internet access (I've got an Imagestream
Enterprise router, a Max 4000, a Max 4002, a Max 4048, and a Max 6000)
knows anything about IPv6.  Even if they did, none of the people who dial
in have the capability to run IPv6 over PPP. To get 6bone access, they'd
have to set up their own tunnels.  This is a pain in the neck for people
who have dynamically-allocated IPv4 addresses, which constitute the bulk
of my customer base.

> > I believe that many people will want to have multiple globally-unique
> > interface identifiers on the same computer.  Is there a standard way of
> > doing this?
 
> DNS. It's a globally unique (and that can be guarranteed unique). 

Yes, host names are globally-unique, but they don't indicate the route
needed to get to the destination, but IPv6 addresses are constructed with
the idea that the structure of the prefix would reveal the routing
structure of the Internet.  That means that the amount of routing
information distributed around the various backbone providers can be
reduced without any reduction in the functionality available to the end
user.

Now, I suppose you could use the DNS to distribute the information that
addresses with various prefixes belong to the same physical adapter, but I
would prefer some mechanism (like an ICMP6 AKA message or some such) that
doesn't involve third parties setting up their equipment correctly and
individuals setting up their own DNS servers entail dangers that I don't
like to contemplate.  (Judging by the difficulty the professionals at Time
Warner Communications have in setting up their DNS servers, I don't want
to think about your average "Joe on the street" trying to do it.)

No, it's got be built in to the software and not dependant on the mercy of
a DNS administrator.
-- 
Jonathan Guthrie (jguthrie@brokersys.com)
Brokersys  +281-580-3358   http://www.brokersys.com/
12703 Veterans Memorial #106, Houston, TX  77014, USA