I listened to the 6man WG on- and off-link presentation and responses from the audio feed. I have a comment relating to what Tony Hain said during the presentation. Inappropriately sending data to the default router is not catastrophic, the data will be forwarded by the router to the appropriate destinations. A redirect may even be issued, if supported by the router, to correct this problem. However, inappropriately sending out an NS to resolve an on-link destination can be catastrophic if the router does not respond to the NS (and if others on the link are not physically connected to the node sending the NS) - as presented in the example of an aggregation router deployment in our slides. Tony's idea to time out an NS and send data to the default router seems like a very useful way to convert a catastrophic scenario into a perfectly acceptable one. However, unfortunately neither the currently existing IPv6 ND implementations nor the ND specifications actually follow/specify this behavior. In fact, the SEND (RFC 3756) says in section 4.2.5 (Bogus On-Link Prefix), "If a sending host thinks the prefix is on-link, it will never send a packet for that prefix to the router." This is quite unfortunate, since now we have to be extremely careful to understand EXACTLY when a destination is on- or off-link. On-link determination is no longer just a performance optimization, it's a basic data forwarding correctness issue. - Wes Beebee
-------------------------------------------------------------------- IETF IPv6 working group mailing list ipv6@ietf.org Administrative Requests: https://www1.ietf.org/mailman/listinfo/ipv6 --------------------------------------------------------------------
