I have a real hard time understanding why proponents of the depreciate SL's claim that SL's would require special handling by applications , while unique PI's don't. If an address has limited scope, i.e., there is an address filter somewhere in the network that prohibits global e2e connectivity, then sometimes applications will fail to establish connections using those addresses.
What I fail to see is why the SL prefix (FEC0::/10) is any different than a unicast prefix (say 2001:0400/32). I contend that it isn't the SL unicast prefix that breaks the e2e principle, but the address filter. The only difference I can see is that the SL prefix guarantees that an address filter will be crossed, while the unicast prefix makes no such guarantee. If we accept that these filters will exist, for all unicast prefixes, then apps will have problems when communicating when peers are separated by this filter.
So I don't accept your point below that SL's require special handling by apps, but unique random addresses that provide the same function don't.
What am I missing that leads you to a different conclusion?
Rich
At 07:01 AM 4/3/03 -0500, Margaret Wasserman wrote:
[snip snip snip]
Site-local addresses have properties that are not needed to address these problems, and that cause problems for routing protocols and upper-layer protocols, such as:
- Ambiguity (requires zone ID to disambiguate). - Need to retain site "convexity". - Single level of nesting (you can't have a further access- controlled site within a site). - Special address selection handling required at the IP layer and in applications. - Places the knowledge of private addressing in all implementations, instead of keeping the knowledge of routing boundaries constrained to the infrastructure (routers, firewalls, etc.)
Using random addresses for disconnected sites was proposed as an alternative, but that is not the only alternative. There are, in fact, two superior alternatives that do not require any standardization work by the IPv6 WG (or any other IETF WG):
- Enterprises could use part of their /48 as private addresses and filter at private borders. This is superior to site-locals because:
- It allows nesting of private areas. - The owner of local addresses can be identified. - Non-ambiguous -- if sent outside the local area, they are unreachable and won't point to the wrong network or node. - Doesn't require special handling on end-nodes or by applications, etc.
- Registries could offer private address allocations to individual enterprises/people. This is also superior for the same reasons listed above.
[snip snip snip] Just my thoughts...
Margaret
-------------------------------------------------------------------- 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] --------------------------------------------------------------------
------------------------------------
Richard A. Carlson e-mail: [EMAIL PROTECTED] Network Research Section phone: (630) 252-7289 Argonne National Laboratory fax: (630) 252-4021 9700 Cass Ave. S. Argonne, IL 60439
-------------------------------------------------------------------- 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] --------------------------------------------------------------------
