Hi Erik > > http://tools.ietf.org/html/draft-ietf-roll-rpl-10#section-6 > > Seems like the non-deployed OSPFv1 used lollipop, but since that was still > subject to the S1 < S2 < S3 < S1 issue, this was removed in OSPFv2. > > That might be an issue for Roll; I haven't looked at it closely. > > In any case, OSPFv1 includes the feature where using a number on the stick > of the lollipop will result in other routers telling you the last sequence > number they they have for your LSA. The message flow for 6lowpan-nd > doesn't have such a mechanism, and given that we don't require flooding all > the information to all the routers, it isn't clear that we can actually add such a > thing to 6lowpan-nd. (Without requiring that the AROs be flooded to all of > the 6LRs.) > > Thus I think there is some work needed before we can consider applying > lollipop or any other sequence numbers to 6lowpan-nd.
[Pascal] Let's do it then. You'll find that RPL: 1) uses simple increments (by 1 I mean) 2) emulates OSPFv2 in that it uses the odd space as a straight part. 3) avoids the S1<s2<s3<S1 pb with a window of consistency check Sequences are comparable if they are not too far apart, and when that's so the higher number wins. If we are still wrong, then the stale information will win until the correct information eventually catches up, which happens sooner if the window of consistency is smaller. RPL has a lifetime associates to most resources and the idea is that the lifetime should be smaller than the time it takes to increments the seqnb outside the window of consistency. Sadly the lifetime is not there for the DIO version nb, but that's another story and probably an issue. Neither RPL nor BR is synchronizing dbs. They are merely trying to identify the freshest information. the consequence of being wrong is a temporary outage for the resource associated to the seqnb. > > > [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 HA in MIP is a centralized solution. > A distributed protocol is q very different beast. > [Pascal] Exactly. So getting the previously used seq nb from the wrong node in the distributed world is probably the wrong idea :) RPL did not takes that path. Pascal _______________________________________________ 6lowpan mailing list [email protected] https://www.ietf.org/mailman/listinfo/6lowpan
