Re: Status of <draft-hinden-ipv6-global-local-addr-00.txt>

[Pekka Savola <pekkas@xxxxxxxxxx>]

On Tue, 3 Jun 2003, Andrew White wrote:
> > I hope you're not implying that apps should know the difference between
> > the two?  That would be broken.  The host probably could, though.
> 
> In most situations, an application would not be required to know the
> difference between the addresses, relying on correct behaviour of
> gethostbyname (destination address selection) and source address selection.
> 
> However, certain applications may wish to deliberately operate at a certain
> scope (such as those doing ip address referrals between hosts which have
> local and global addresses and those which only have global addresses) and
> these would need to override the default address selection rules, thus being
> conscious of address scoping.

Yes, applications should of course be *able* to influence these, if they 
desire so, provided that:
 - they don't need to if they don't want to (including multi-party apps)
 - the API is de-facto standardized so every app doesn't have to reinvent 
the wheel for 10 different operating systems

> > I think a better method would be to give preference to global unicast,
> > ie., reverse the scoping rule to prefer the greatest scope.
> 
> On what grounds?  Local scoped addresses are not mandatory, and thus should
> only be put in place if they are intended to be used.

Yes, but used by *WHOM* ?

If they're used in the site, and that only, typically that would be OK.

This also begs the question whether local-scoped addresses would be
deployed without global addresses or not (as some have required).  If
there are local-only nodes, the reason to deploy local-scoped addresses
could be simply to use them when needed, no more no less -- without 
disturbing the global communication of nodes with global addresses!

> To my way of
> thinking, the presence of a locally scoped address suggests that the network
> is unwilling to trust the global address for local communication, 

The motivations vary a lot.

> and would
> thus expect the local address to be preferred for local-local traffic.

That may be a desire, yes, but they may be others -- particularly 
regarding how you intend to deploy the addresses.

To me, by default using the global addresses would seem like a plus: even
if/when they leak (to a site which is also using local addresses), nothing
bad happens: the global addresses are still tried first (and not second)
-- the way it should be.  This way, people could also publish local
addresses in the non-split DNS (whether that's a good or bad thing is a
different issue).

> > > However it is useful if a host can have both types of addresses
> > > and use them appropriately.  This creates a host with multiple global
> > > addresses, a form of multihoming.
> > 
> > I fail to see how a local and a global address could be considered a form
> > of multihoming.
> 
> It depends on your definition of multi-homing.  I've hear the term used in
> the following ways:
> 
> - A network with access to the global internet via more than one independent
> path.

In some cases, it's more detailed than that, but yes.

> - A node with more than one usable address (having addresses in more than
> one logical subnet simultaneously).

This is an abuse of the term.
 
> The second definition is being referred to above.

That's "multi-addressing".  (Note that there's a significant overlap with 
the two definitions above.)

-- 
Pekka Savola                 "You each name yourselves king, yet the
Netcore Oy                    kingdom bleeds."
Systems. Networks. Security. -- George R.R. Martin: A Clash of Kings

--------------------------------------------------------------------
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 majordomo@sunroof.eng.sun.com
--------------------------------------------------------------------