> I agree with Hesham.  We shouldn't be applying "MUST NOT"s to situations
> where we have workable solutions.

the closest thing we have to a workable solution is to find a way to
give every site that connects to another network a global prefix 
(whether it's globally connected or not) and have applications 
ignore the SLs (at least for purpose of referrals).  and if you do 
that, there is little point in having SLs at all except for
isolated networks - they just complicate management and DNS etc.
for the networks that use them.

> > actually, no.  people have shown that there are ways of
> > using SLs for some apps in such situations.  nobody has
> > come up with a general solution that requires less than
> > either:
> > - expecting all non-isolated networks to provide global
> >   addresses (and having apps ignore SLs in the presence
> >   of globals), or
> > - expecting apps to do their own addressing and routing.
> 
> Which is just fine!  Nobody was arguing that non-isolated networks
> shouldn't provide global addresses -- they should.  If someone want to
> talk globally, they should use global addresses.  

but this isn't just about sites that want to talk globally - this
is about sites that connect to other sites, some of which talk 
globally.   and again, if any of those sites use SL, then apps
that communicate across site boundaries are forced to implement
their own addressing and routing.

> (I'd make a small argument about your "ignore SLs in
> the presence of globals", however, since I already showed how using
> site-locals can work if the site-local is never passed outside of its
> scope zone).

yes, but that presumes either that all such sites have global addresses
or that there is never a need for an extra-site party to perform a
referral between two hosts on the same site.  neither seems reasonable.

> I claim that requiring non-isolated networks to provide global addresses
> to those entities which desire global connectivity *is* a general
> solution 

I disagree.  network A isn't necessarily providing global connectivity 
(transit) to network B even though A as global connectivity and
A and B connect to one another.  

the problem is that applications that communicate between A and B
and perhaps with nodes on other networks need a linear address
space, and they don't have one if any of those networks use site-locals.

also, insisting that A give up a part of its address space to B
just in order to connect to B can be problematic - it affects A's
existing organization of its address space.  (just because so much
address space is available does not mean it will be well-utilized!)
my guess is that enterprise networks will be reluctant to give up
part of their address space to allow other enterprises to connect.
and if B wants to connect to several networks, which one gives up 
part of its address space?

Keith
--------------------------------------------------------------------
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