I'm very much opposed to extending RA for this purpose. Having 2 ways to do the same thing serves to increase complexity of the overall ecosystem. Unless you can guarantee that the 2nd method will only ever be used in a closed system (network manager consciously knows that this is the mechanism all installed devices support and will use, so that interoperability is guaranteed). If you cannot guarantee that it will only exist in a closed system and that no user will ever be confused, then by introducing a 2nd method, you are increasing complexity (more than doubling it, for this function) for the majority of devices (that will have to implement both to be assured of interoperability, and will have to implement "tie-breaker" functionality as well). Doubled complexity for the majority in order to achieve increased simplicity for a minority is a Bad Idea.
Whenever there are 2 ways to do the same thing, decisions must be made regarding the positioning of both methods in the marketplace. Specific to this proposal, the following scenarios are possible outcomes: 1. Clients that consumers (residential, small, and medium business) expect to interoperate in an open (not single vendor, no "check for special restrictions") environment are unpredictable and will do one or the other. Therefore, routers that expect to interoperate in an open environment must do both. This outcome successfully doubles the complexity of providing NTP info via a router, because the router must support both DHCP and RA methods. This happened with DNS. Most routers are supporting both ways. Ugh. But it is what it is. So with DNS the router people bit the bullet and did double the complexity. 2. Routers that consumers expect to interoperate in an open environment are unpredictable and will do one or the other. Therefore clients that expect to interoperate in an open environment must do both. This outcome successfully doubles the complexity of a client who needs to get NTP. Also a bad thing. 3. The people who ask for the special RA capability say that the RA capability is only for use in special closed environments, and consumers will never find themselves confused by trying to use these devices in an open environment. Really, it will be ok, and it will never morph into outcome 1 or 2 above. Or outcome 4 below. We promise. 4. [my favorite] Neither clients nor routers can predict whether they will find themselves in an environment where only one or the other is supported. So both clients and routers that expect to be fully interoperable must support both. The majority gets to double their complexity. So that a minority can simplify just a little bit. Barbara From: [email protected] [mailto:[email protected]] On Behalf Of Roger Jørgensen Sent: Sunday, December 25, 2011 9:06 AM To: Doug Barton Cc: 6man Mailing List; Brian E Carpenter Subject: Re: IPv6 Router Advertisement Option for Foobar Configuration On Dec 23, 2011 8:57 AM, "Doug Barton" <[email protected]> wrote: <snip> > What I *am* saying is that extending RA is always the *wrong* answer. Yes it is. Keep RA simple and lets solve "problems" elsewhere, dhcp is the current tool. --- Roger J ---
-------------------------------------------------------------------- IETF IPv6 working group mailing list [email protected] Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6 --------------------------------------------------------------------
