[Working group chair hat off]

A few comments on the Site-Local discussion that I did not see getting discussed or proposed.

There was a reference made to networking airplanes somewhere in this thread. If my memory is correct, the airplane industry did select an open standard for airlines. Some planes run ISO CLNP over FDDI. They did pick an open standard, but.....

A lot of the discussion seems to imply that site-local addresses are created automatically (like link-local). This is, of course, not the case. Site-local are only created if someone configures a router with a specific site-local address on an interface and tell the router to advertise the prefix in a routing protocol to other routers and to advertise it to hosts on the link in RAs. Site-local addresses will only appear in a site if an administrator decided to configure them.

It seems to me that from a reachability point of view there isn't too much difference between using site-local addresses and having a firewall that blocks traffic from one set of global addresses and another. Firewalls create limited scope addresses from global addresses. If one assumes the existence of IPv6 firewalls, then these firewalls can also enforce site boundaries. Firewalls are essentially doing this today. Compared to what firewalls already do today, this is trivial. If the site boundaries are going to be defined by firewalls, then it is probably less important that we have to define multi-site router behavior.

One of the issues that was discussed regarding the ease or difficulty of configure site-local scope filters in routers. It seems to me that the simple way of doing this is configure the router with the zone that each interface is in. The router would then automatically create internal filters that enforce the site boundaries. This seems much easier to me than having to create filter rules based on the prefixes that are assigned to each interface.

Another router issue that gets talked around is should packets with site-local destination be forwarded to "default". Given that site-local addresses are not created without being configured, one approach could be to have a "black hole" route for FEC0::/10 preconfigured in all routers. The router would then only forward packets with site-local to destinations that matched more specific routes. They would never get forwarded to the ISP via the normal default route.

From a private exchange with someone from a router vendor. They are taking the approach of creating a "no-site" site in their product. That is if an interface configured to be in the "no-site" site, the router will not forward any packets with site-local addresses to/from this interface. This might make for a simple default behavior site border router.

Bob


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