A good summary but let me add a few points.
The Microsoft IPv6 implementation supports Erik's draft. It also puts
site-local addresses in the DNS. And in normal home gateway scenarios
(requires some administrative action to stop this) it will send RAs with
a site-local prefix as well as a global prefix.
Suppose a host X, not implementing Erik's draft, resolves a DNS name and
gets back a global and a site-local address. If host X does not have a
site-local address itself, and host X implements
draft-ietf-ipngwg-default-addr-select-02, then the name resolver
function will return the global address before the site-local address.
So host X will use the right address.
Suppose host X does not implement Erik's draft and does not implement
draft-ietf-ipngwg-default-addr-select-02. Perhaps the name resolver
function always returns addresses in the order it gets them from DNS.
Then it's unpredictable which address host X will use. If it uses the
site-local address it will probably have a lengthy timeout before
falling back to the global address, but in the worst case the
application/user might give up, or it might even end up communicating
with the wrong host.
Another possibility is that a DNS server can look at the source address
of the DNS request. If the source adress is an IPv6 site-local address,
then the DNS server can include site-local addresses in its reply. If
the source address is an IPv6 global address, or an IPv4 address, then
the DNS server can filter out site-local addresses. This introduces the
constraint that a site has to contain a DNS server to use site-local
addresses. This technique could be used in conjunction with Erik's
draft.
Rich
> -----Original Message-----
> From: Paul Francis [mailto:[EMAIL PROTECTED]]
> Sent: Thursday, February 15, 2001 1:38 AM
> To: Robert Elz; Brian E Carpenter
> Cc: [EMAIL PROTECTED]
> Subject: summary of Re: another renumbering question
>
>
>
> Just an attempt to summarize this thread:
>
> Jim Bound thinks the use of site-local addresses is bad.
>
> Robert Elz things that the use of site-local addresses is ok, but that
> two-faced DNS is bad.
>
> Other folks don't think two-faced DNS is so bad. (I'm in
> this camp too...)
>
> Robert Elz also made a good suggestion as to how hosts can learn the
> site-local addresses of other hosts in the site (least-wise
> it seemed good
> to me). But the state of this suggestion is that it exists
> in an email and
> nowhere else (i.e., no internet draft), so unless someone writes it up
> nothing will happen to it? (By the way Robert, are the ICMP
> "tell me your
> site local addresses" and ICMP "site local source exiting the
> site" messages
> you mention in your suggestion in existing drafts, or are
> these new messages
> you are making up?)
>
> In the absense of Robert's suggestion or Erik's draft
> (draft-ietf-ipngwg-site-prefixes-05.txt) (both of which
> require changes to
> existing implementations), I gather that there are only two
> possibilities:
>
> 1. Don't use site-local addresses at all.
> 2. Use site-local addresses and two-faced DNS.
>
> Is this a reasonable summary?
>
> PF
>
>
>
> --------------------------------------------------------------------
> 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]
> --------------------------------------------------------------------
>
--------------------------------------------------------------------
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]
--------------------------------------------------------------------
