Leif Johansson wrote: > Unfortunately there does not seem to be a hypertext archive > of the list but the post was from 2003-07-08:
The manual spam filter must have fat-fingered that one into the trash ... > > =============== paf =========== > > I don't know how to attack the people which talk about "intermittent > connected networks". > > The problem I see is that people are locked into the IPv4 > world where a > DSL modem have one address which might change every connect. That is > extrapolated into one IPv6 prefix which changes every connect. > > Instead, to me, the user should get the _same_ IPv6 prefix > every time he > connects. As long as one connect to the same upstream > network, the IPv6 > prefix should not have to change. I believe that most service providers are looking at this as their long term goal, but as noted there are conflicts with their IPv4 model. > And, switching upstream network will > force a switch in prefixes of course. The renumbering event > comes when > _connecting_, not disconnecting and connecting as the address can be > kept until a new prefix is to be used. > > Is this not true? If there is only one connection to any other network, it is possible for a disconnected network to continue using a prefix for some time. If there is a connection to another network and the prefix were reassigned, there is a clear conflict that needs to be resolved. In any case, this presumes that the network will always reconnect to the same SP, and that it is ok to disrupt local communications whenever a connect event happens. Neither of which are valid for all networks. > > Further, we do not have a good session layer in the IETF > which means a > network _always_ have to reset the connections when a renumbering > happens, which is the same as a more drastic network topology change > happens. The solution to this is mobile IP. See my note from yesterday about 'solving the real problem'. Mobile IP is a hack to create the illusion of stability at L3. Even if we decided that was the answer, what it means is that every local connection is created as a mobile IP connection, because there is no guarantee that the routed prefix won't change do to external events. While we can do that technically, there will be many who complain about the performance hit of having to pad out every local packet with a home address. The application layer wants to see stability, and that is fine. The transport layer and below deal with the reality of shifting topology. What we need is a formal layer between transport & applications that provides the illusion of stability up, while managing the reality of constant changes below. I specifically used the term stabilization layer, because 'session' appears to be a loaded term. I really don't care what we call it, but it appears to logically exist between layers 4 & 7, because that is where the demands/reality differ. Tony -------------------------------------------------------------------- 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] --------------------------------------------------------------------
