Hi, Karl Thanks for your reply, now I understand your arguments clearly.
I think our common view is "ambiguity in not good". Then your way is just deprecating them; if don't deprecate, then make it clear. But I just have the opposite approach: since it's ambiguous, we need at least make them clear, and beyond, we need more SLAAC/DHCPv6 interaction than the original M/O flags semantics. Let me explain what are "the current problems": [1]. Renumbering problems. In RFC5887(Renumbering still needs work), section 5.1.1 says clearly: "Until this ambiguous behaviour is clearly resolved by the IETF, operational problems are to be expected, since different host operating systems have taken different approaches. This makes it difficult for a site network manager to configure systems in such a way that all hosts boot in a consistent way. Hosts will start SLAAC, if so directed by appropriately configured RA messages. However, if one operating system also starts a DHCPv6 client by default, and another one starts it only when it receives the M bit, systematic address management is impeded." Our test had identified the host OSes have tanken different apporoaches. For the above mentioned "systematic address management" , we already discussed several instance, as I described in the draft: - The site manager wants the host to do DHCPv6 when online. Current practise is sending RA M=1 and don't include PIO, however, how the OSes will behave is not controlled by the site. - The site manager wants the SLAAC-configured hosts to do DHCPv6 (e.g. adding a multihomed uplink who uses DHCPv6 for address configuration). For some OSes, there's just no availeble method. - The site manager wants the DHCPv6-configured hosts switch to SLAAC. If you don't interprete M=0 as prescriptive, the only way is either just waiting for the DHCPv6 lease time expires, or initial DHCPv6-reocnfiguration from the server side. But I don't think it is an efficient way when renumbering. If the lease time is too long, the problem is more serious. For DHCPv6-reocnfiguration, it is only unicast, and stateful, not proper for bulk use. [2] Requirement from ISP ISPs have strong requirement of strictly controlling the CPE behaving as they expect. There's an instance I learned, an ISP deployed stateless DHCPv6 in their IPv6 network, which means they required the CPE to configure addresses through SLAAC and other parameters through DHCPv6. Since the ambiguity in standard, they had to directly specify the requirements to the CPE vendor, which means, universal CPE may not be availeble in IPv6 networks unless we have clear definition in standard. [3] I think Alexandru's argument of mobile nodes is another good example. Regards, Bing ________________________________________ From: ipv6-boun...@ietf.org [ipv6-boun...@ietf.org] on behalf of Karl Auer [ka...@biplane.com.au] Sent: Wednesday, August 01, 2012 17:08 To: IETF IPv6; re...@ietf.org Subject: RE: about DHCPv6/SLAAC interaction On Wed, 2012-08-01 at 20:35 +0000, Liubing (Leo) wrote: > May I venture to argue that, "had been discussed before" might not be > reasonable enough to deal with current problems. What exactly are "the current problems"? It seems to me there are only two: - it is not explicit whether M means a host "must", "should" or "may" use DHCPv6 to obtain an address. It is fairly clear in practice that this is a "should" - it is not explicit that the M and O flags are completely independent of each other. Common practice is to treat them as independent, and I can see no good reason to link them in the RFC or anywhere else. - there is a question about whether a host obtaining a DHCPv6 address should NOT perform SLAAC. Since the RFC certainly doesn't say it should, it is a mystery to me why people think DHCPv6 excludes SLAAC. Access to addresses via DHCPv6 can be completely controlled on a subnet by subnet basis in the DHCP server, and SLAAC can be completely controlled on a subnet by subnet basis using the autoconf flag in RAs. - if I were King of the World I would deprecate both flags. They serve no useful purpose. > this real-network problem may be much more sensitive than we > imagined/discussed to be when writing RFC4862. Again, I don't see the issue(s) here. > So, my personal preference is similar with Karl's, we might need > additional flags to cover the SLAAC/DHCPv6 interaction semantics. That is NOT my preference. My preference is to deprecate both flags. My second preference is to clarify as above. All that stuff about additional flags was thinking aloud about how M/O could be made mandatory; it was certainly NOT a suggestion that they *should* be made mandatory. > And in 6renum, we also discussed some new semantics requirements (see > in the draft, and the section 5.1 in 6renum WG item: > http://tools.ietf.org/html/draft-ietf-6renum-gap-analysis-02 ), I've just read that doc and as far as DHCP and SLAAC go, it seems to be composed mostly of questions. I really think the draft should state what the *problems* are that it seeks to solve. Regards, K. -- ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Karl Auer (ka...@biplane.com.au) http://www.biplane.com.au/kauer http://www.biplane.com.au/blog GPG fingerprint: AE1D 4868 6420 AD9A A698 5251 1699 7B78 4EEE 6017 Old fingerprint: DA41 51B1 1481 16E1 F7E2 B2E9 3007 14ED 5736 F687 -------------------------------------------------------------------- IETF IPv6 working group mailing list ipv6@ietf.org Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6 -------------------------------------------------------------------- -------------------------------------------------------------------- IETF IPv6 working group mailing list ipv6@ietf.org Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6 --------------------------------------------------------------------