On Tue, 2002-11-26 at 16:43, Keith Moore wrote:
> > As I understand it, the GUPI model pretty much gets rid of the site
> > model completely - it really is just catagorising connectivity and
> > destinations as internal verses external - internal connectivity using
> > GUPI addressing, via internally administered network infrastructure
> > (including a VPN), verses external connectivity via the public Internet,
> > using global addressing.
> 
> I wouldn't use the words "internal" and "external" because GUPIs would
> surely be used for private connectivity between sites.  I'd use
> "private" and "public" instead.  So GUPIs would be used within "private"
> sites that (for the most part) didn't have connectivity to the public 
> Internet, and for connectivity between "private" sites and other sites;
> globals would be used by sites with connecitons to the public Internet.
> 
> Keith

I think your imagined model of how GUPIs would be used might be
different to mine.

As I see it, there are only three basic addressing scenarios, assuming a
network with more than one segment (ie using link local addresses for
internal connectivity is not covered here) :

1) Globals - the address space stability of the global address space
given by your provider is good enough. You use globals for everything,
and have decided that coping with a global renumbering event is cheaper
than the administration of GUPIs. 

2) Globals and GUPIs - you don't want to rely on the stability of your
allocated globals for your internal connectivity, so you roll out GUPI
address space as well. GUPIs are used for your internal communications
ie communications that doesn't travel across links that are part of the
public Internet.

3) GUPIs only - you don't have a Internet connection, you need internal
address space, GUPIs suit. If at a later date you get an Internet
connection, you leave your GUPI address space in place, and roll out
your provider allocated globals.

A network with occasional public Internet connectivity would switch
between scenario 2 and 3, based on the presence of its Internet
connection.

My comments about naming (and pretty much all others on GUPIs) have been
based on scenario 2. I think scenario 2 would be the most common
encountered.

In the conext of my above scenarios, I'd be a bit concerned that the
word "private" has negative IPv4 NAT associations - using it with IPv6
may help people along the path of thinking of using GUPIs and IPv6 NAT.
The word "private" also suggests security, which we don't want to
further endorse. 

I think that "internal" and "external" are good description words of
what types of communications is happening in scenario 2.

Regards,
Mark.



--------------------------------------------------------------------
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