In your previous mail you wrote: >I have two concerns about this: > - the random order is not defined (this is a formal concern because > the intention is clear) I think most people will understand the intent of "random order".
=> as I've written this is a *formal* concern... (formal = on the form) > - we moved from an implementation hint to a SHOULD, this will make > old implementations not conformant and in some situations there > can be far better solutions. So I prefer to get a MAY (enough to > get this proposal implemented in the future by everybody without > harm) and an explicit reference to the default router preference > draft (first because I am in favor of this from the beginning, > i.e. since many years, second because this is the best reply to > the "special case" argument (not so special case because 100% of > the IPv6 networks I used have several routers with a better one)). I disagree. The intent of the document is to change the behavior from a hint to a common behavior. Hence the SHOULD. => if this is the intent the document is underspecified (i.e. it have to remember when the default router selection is done (this is at the beginning of 6.3.6)). There is too a real issue about the required minimal default router list length (two issues in fact: do we allow small implementations to use a short list (possibly shorter than the real number of default routers)? What is the reply to the bad guy who tries to give us 2^64 default routers?) Note there is something about this in 5.3 GC and Timeouts and 6.3.4 Processing rcvd RAs. Note that the current "hint" for round robin is less than perfect. => I agree but my concern is more about the requirement level. I think the current implementations are covered by the SHOULD. => they are not compliant with the SHOULD, not a real problem for round robin (as it can be looked at a bad random order) but implementations which keep the same router have a very different behavior (than the recommended one). If it was a MUST there might be a problem with current behavior. => I agree but a SHOULD in a draft standard (i.e. the intent) is really quite strong. In Salt Lake City where the issue of combining this with the default router preferences was discussed, the consensus was to keep this separate. => my concern is there are a lot of situations where some potential default routers are better than others. Without the default router preferences the only way to control the default router choice is to use zero router lifetimes. This is far too drastic and can't express any kind of nuance... So I'd really like to get preferences before or at the same time than random selection. The plan was to also add this behavior to the default router preferences draft for the case where there are multiple routers at the same preference. => we agree but there is clearly a problem in priorities. Regards [EMAIL PROTECTED] -------------------------------------------------------------------- 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] --------------------------------------------------------------------
