> > On 06/28/10 09:04 AM, Pascal Thubert (pthubert) wrote: > > > The sequence number, that was provided in earlier versions of the ND > > spec, is required to resolve movement from double registration in > > both RPL and the backbone router spec. It makes more sense to me to > > have the sequence number in the ARO right away than to have to make it > > an option in case the network grows past what's acceptable without > > coordinated LBRs. > > > > What do you think? > > The issue with requiring hosts to send a sequence number in ARO is that it > also means requiring the hosts to have some stable storage such as flash > where they can store the most recently used sequence number. > > Otherwise, if a host registered with sequence number 0x12345678 and then > power cycles and looses that state, it might try to register with 0x00001234, > which will be ignored by the routers since it would be considered too old. > > I know earlier versions of the draft had this initial set of numbers that were > reserved for "I've lost state - please accept this". But that doesn't seem to be > robust.
[Pascal] This is a very classical issue. RPL handles it with the also classical lollipop algorithm. And yes, 2 reboots in row will take a bit more time to settle than just one, but it ends up working after the second reboot nb reaches the previous watermark. More on the RPL version at: http://tools.ietf.org/html/draft-ietf-roll-rpl-10#section-6 > Suppose a host registers using sequence number 1. Then it moves and re- > registers using sequence number 2. However, a delayed registration packet > with sequence number 1 is propagating through the network. When it > reaches the 6LBR it will win over the current #2, because it it is in the "I've lost > state" range of sequence numbers. > > I don't know of a robust way to handle this *without stable storage on each > host*, using a single packet. It could be done with a 3-way exchange where > the router would send back "is 1 your current sequence number", and get > back an ACK/NAK from the host. But that is a fair bit more complicated, > especially if we apply this to multi-hop DAD. > [Pascal] This is basically what the HA does in MIP. This has an added value for the HA to be able to ensure bidir reachability by using a new seq num as a challenge, though it is not officially used in the protocol. I agree with all the above and favor the lollipop for those reasons. Pascal _______________________________________________ 6lowpan mailing list [email protected] https://www.ietf.org/mailman/listinfo/6lowpan
