Hello again,
JINMEI Tatuya / 神明達哉 wrote: > But, I'm okay if the ipv6 wg (in this case) will discuss if an > extension mandating something to all nodes is acceptable, and the wg > can even start with rejecting former proposals which are MUST. > (here I'm just talking about the general procedure, rather than the > particular issue of Home Address Option). I reckon that before IETF last call on any universal IPv6 mandate, the mobile-ip working group will have the responsibility to make the reasons for the mandate clear. >> What do you think about the points I tried to make a few days ago >> about general scalability and the likely proportion of future nodes >> that will be mobile nodes. > > I respect your insight, but my honest feeling is that it was a guess. It's not a guess that reverse tunneling lengthens the response time for packets sent from a mobile node. That's just arithmetic. In my previous note, I clearly labeled the facts and opinions. Did you dispute the points I listed as facts? Did you dispute the points I labeled as opinions? > We do not have actual results about the delay when communicating > without route optimization. This, again, is arithmetic. Plus, there have been publications carrying out the arithmetic and simulations, albeit from IPv4 route optimization. The difference between IPv4 and IPv6 does not matter here. > We're (or I am) not sure about the real > rate of nomadic nodes *speaking mobile-ip* in the future Internet. > IMO, requiring a MUST (for all nodes) due to a performance issue needs > a concrete evidence like results of performance tests, not a guess. I'd be very interested to hear how you can resolve the following problem: - If we do NOT mandate the required feature, and a lot of millions of mobile nodes implement Mobile IPv6, then there will be a lot of millions of IPv6 nodes built between now and then that will not work well. - If we DO mandate the required feature, and Mobile IPv6 is a flop, then there will be a lot (but probably somewhat fewer than above) nodes that "unnecessarily" implement the feature(s). - If we mandate the feature LATER, then in the meantime and for a year afterwards, many of the devices manufactured to be compliant with IPv6 specifications (that is, _many more_ than we have now) will suddenly become noncompliant. The intended effect is to reduce the long-term pain of noncompliance. If we settle on SHOULD, then I think that we should also agree to periodically re-evaluate the evolution of mobile networking in the IPv6 Internet, and also to make it clear to manufacturers of all IPv6 devices that noncompliance entails peformance penalties for all communications with mobile devices (that implement Mobile IPv6). A manufacturer that fully understands the trade-offs, and believes that the applications running on their device do not have any special needs for faster performance for streams to/from mobile devices, would then take the option to streamline their implementation. This would be reasonable. The approach of interpreting SHOULD as a license to speed up product introduction is the problem I am mostly worried about, and many people express concern that SHOULD is taken with the unfortunate meaning instead of the literal meaning from RFC 2119. > I recall the discussion about "minimum requirements" for low-cost > appliances or for 3G devices. The authors of the drafts proposed to > omit some features which are mandated in existing RFCs, due to limited > resources and usages. The result of the discussion was "we cannot > allow such non-compliant implementations that easily". Once we > mandate route optimization, the same situation may happen in the > future. I thought the discussion about "minimum requirements" was pretty good! In fact, it's just the kind of discussion that was needed, as shown by recent conclusions about NUD. I also think that similar analysis is needed by vendors before they decide whether or not to implement the route optimization features. > Since such low-cost devices will also be in a large part of the future > Internet (v6) (though some of such low-cost devices can act as a > frequent correspondent nodes), I don't think it a good idea to add > another restriction at this early stage. In theory, we can change > requirements described in RFCs if we find them inappropriate in the > future. However, the recent discussion about the low-cost devices > shows it will be very hard in practice. It is important to keep IPv6 implementable for low-cost devices. This has been a driving force behind the design of HoTI, CoTI, HoT, and CoT, as well as HoA. These are expressly intended to be light-weight signals. I feel that you are making the discussion in the abstract, instead of determining exactly how much harder it is to do the route optimization messaging. Furthermore, the analysis should be taken with respect to the _incremental_ cost, once IPsec has already been implemented. >> ... nodes secured by IPv6 IPsec AH occupy only a small portion >> of the Internet. Do you apply the same argument to convince yourself >> that the IPsec mandate was a mistake? > > No, because in this case it's an all-or-nothing choice; if we do not > implement IPsec, we cannot be secure (at least at the IP level). Actually, I think it's exactly the same from THIS argument. A node that does not implement IPsec, cannot run secure applications. A node that does not implement route optimization, is far less likely to be able to meet delay bounds, and also will experience some (perhaps quite significant) loss of robustness in communications. > On the other hand, even if we do not implement the route optimization, we > can still communicate with mobile nodes in some less-efficient manner. > Of course, one can say that the mobility case is also an > all-or-nothing choice, i.e. that "the less-efficient manner" means not > working. But, at least to me, the latter argument is not as clear as > the former - again, we do not have concrete results about the > performance. No, it is just as you have described it. But, for some applications, performance matters. You wouldn't watch a TV show with insufficient bandwidth to your device, I'm sure. You wouldn't use a handset for voice applications if handover performance stunk. To be precise, mandating route optimization will enlarge the realm of applicability for deployment of applications on IPv6 mobile devices. Not doing so means you'll have to find other protocols to run those applications. Full employment! Regards, Charlie P. -------------------------------------------------------------------- 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] --------------------------------------------------------------------
