Unless anyone can mention a lot of potential IPv6 deployments that will deploy IPv6 without a router sending RA, it's not even worth discussing the subject. Don't tell me one will deploy DHCPv6 hosts without a router in the deployment - don't give me private closed networks for an example - it's a weak argument. Even a dinky home network I can build in my home for IPv6 better have the Linksys router sending RA's to IPv6 hosts in my home. If a router is deployed, then why disable the RA on the router? I agree with James that such a RA-absent network is plain broken.
My union statement has been qualified by "if a network has RA and DHCPv6, both, sending prefix lengths". If such a scenario happens, then what choice does one have but to resolve discrepancies in data provided by two sources but to use some sort of union. >From day one, I am with James and Bernie, that DHCPv6 should not add any more options towards prefix information or gateway. Let only RA provide prefix information. Hemant -----Original Message----- From: James Carlson [mailto:[EMAIL PROTECTED] Sent: Friday, August 17, 2007 12:34 PM To: Hemant Singh (shemant) Cc: Templin, Fred L; Iljitsch van Beijnum; [EMAIL PROTECTED]; ipv6@ietf.org; JINMEI Tatuya / ???? Subject: RE: [dhcwg] Re: prefix length determination for DHCPv6 Hemant Singh (shemant) writes: > Thanks, James. I agree with Fred then that a node can try DHCPv6. But > now how does the node get a prefix length? As you are saying, some > manual or static configuration can be used. I certainly don't like the > host to assume any prefix length in this scenario. Since I am not a > fan of any manual configuration, it does make sense, only for such a > case of absence of an RA, that DHCPv6 provides prefix length. Since > DHCPv6 doesn't know if the network's router will issue RA's or not, > then DHCPv6 has to provide prefix length all the time. My view on it is that such a network is just plain broken. V6 seems intentionally designed to assume that networks have routers, because routers are the _authoritative_ source of prefixes (they have to route them). Thus, it's a case I test for, and one I make sure "works," but I assume that it's useless. In my own implementation, you end up with a /128 for that interface. If you have a default route, everything will go there (via the link-locals), and you'll get a pile of redirects for the local hosts. If you don't, then you're just sunk. Enjoy the address, because it's all you've got. ;-} > Then I am for what Iljitsch is saying. If a host see a discrepancy in > prefix lengths from RA and DHCPv6, then host has to decide based on a > union of information. "Union" doesn't make much sense to me. That just provides a needless opportunity for error, and consequent bizarre and hard-to-diagnose behavior out of subsets of hosts on a network that get the wrong data. -- James Carlson, Solaris Networking <[EMAIL PROTECTED]> Sun Microsystems / 1 Network Drive 71.232W Vox +1 781 442 2084 MS UBUR02-212 / Burlington MA 01803-2757 42.496N Fax +1 781 442 1677 -------------------------------------------------------------------- IETF IPv6 working group mailing list ipv6@ietf.org Administrative Requests: https://www1.ietf.org/mailman/listinfo/ipv6 --------------------------------------------------------------------