I agree. jak
----- Original Message ----- From: <JINMEI Tatuya / [EMAIL PROTECTED]@C#:H (B <[EMAIL PROTECTED]>)> To: "James Kempf" <[EMAIL PROTECTED]> Cc: "Pekka Nikander" <[EMAIL PROTECTED]>; "Nick 'Sharkey' Moore" <[EMAIL PROTECTED]>; <[EMAIL PROTECTED]>; <[EMAIL PROTECTED]> Sent: Monday, March 01, 2004 1:28 AM Subject: Re: SEND/DIID interoperabilility (was: Re: [rfc2462bis issue 275] DAD text inconsistencies) > >>>>> On Mon, 1 Mar 2004 00:12:11 -0800, > >>>>> "James Kempf" <[EMAIL PROTECTED]> said: > > >> Putting that aside, a SEND node could well *defend* the address > >> fe80::A for DAD/DIID purposes, but it would never actually use it. > > > I think that's the issue. Should a SEND or 3041 node be required to defend > > LL addresses that use suffixes configured for their global unicast addresses > > even though they will never use the LL addresses? > > > Since it's a backward compatibility issue, I think it would depend on how > > widely implemented DIID is. If it's not widely implemented, then there's no > > point in doing it. My personal feeling is that it is an extra bit of > > signaling that is unnecessary unless there is a widely deployed IPv6 capable > > OS out there that does DIID. > > > In any event, the SEND spec doesn't need to change, because this behavior > > would have to be specified in RFC 2462bis since it would apply to 3041 nodes > > too or for that matter any node that used a different algorithm for > > configuring its suffix. > > I agree on all the points. And my conclusion still does not change: > I don't see the need for any change on this in rfc2462bis. > > I don't know the deployment status of the DIID implementation, and we > should be careful not to underestimate it. However, I think we should > also consider the reality of the problematic case from an operational > point of view. > > Recall the scenario I described again. The problem occurs when the > DIID node *happens to* configure fe80::A and P::A. But can this > really occur in a practical sense? In this particular case, "A" is a > CGA identifier, so it cannot equal to a normal EUI-64 identifier. > Also, "A" would look like a non-readable, random 64-bit number, so it > should be very unlikely (in a practical sense) for the DIID node's > administrator to use it as a manually assigned interface identifier. > > A similar argument should apply to the case of RFC3041. > > Meanwhile, if we agree on the proposed change in rfc2462bis (MUST do > always DAD for each unicast address), new implementations will simply > follow the requirement. Also, existing DIID implementations will > gradually migrate to rfc2462bis. So, the actual odds to have the > problem will get smaller and smaller (starting with very small odds). > > It should also be noted that one of the key motivations of the > proposed change in rfc2462bis is to simplify the specification and > implementation. If we introduce an additional defence algorithm to > avoid the "compatibility" issue, the resulting specification will > relatively complicated, degrading the effort to simply it. > > To summarize the points, I'd propose to do nothing on this in > rfc2462bis because: > > - the actual odds of the problem in the real world should be very low, > even in the current status. > - if we take the proposed change in rfc2462bis, the odds will be > getting lower and lower. > - if we add something for this to rfc2462bis, the protocol will become > complicated, degrading one of the motivations of this clarification. > > JINMEI, Tatuya > Communication Platform Lab. > Corporate R&D Center, Toshiba Corp. > [EMAIL PROTECTED] > -------------------------------------------------------------------- IETF IPv6 working group mailing list [EMAIL PROTECTED] Administrative Requests: https://www1.ietf.org/mailman/listinfo/ipv6 --------------------------------------------------------------------