Hi Jonathan: >On Jun 11, 2009, at 5:53 AM, Pascal Thubert (pthubert) wrote: >> Looks so alike and still is so fundamentally different. Ralph >> explained >> that in SFO a lot better than I can ever do but that has to do with >> the >> model below. >> >> In DHCP, the server owns the address and lends it away. You have to >> get >> back to that server to renew the binding. In whiteboard, the nodes >> owns >> the address and registers it where it wants, or anywhere (anycast). >> >> When we do SeND, that distinction might blur quite a bit but still, >> the >> white board acts on behalf of the node so it does not hold any master >> state (like a pool with LRI etc...) after the node is gone. > >Hi Pascal, > >If Ralph could reiterate what he said at the WG meeting on the ML, >that would certainly help my understanding of the fundamental >difference between DHCP and whiteboard. > >At least in my mental model, the whiteboard is authoritative in what >nodes can use what address. The node MUST periodically renew the >binding with the whiteboard. The node cannot use that address when the >binding expires without renewal because the whiteboard could then >allow another node to use that same address. Whether or not we view >the node as "owning" the address is a non-issue for me. Functionally, >the whiteboard is authoritative and that's not unlike DHCP.
Seems my words failed to convey the message and I hope Ralph will express that better. The key in my mind is that the owner is the node, and the whiteboard is just an attorney. Though he is the one who speaks during trial, the attorney can only say what his client agreed upon. And the client might switch attorney at will. With stateful DHCP, the server plays is like a chess player that places and controls its pawns at will. So the model is reversed. >Also, in various places in the ND draft, we say *stateless* address >autoconfiguration - when in fact this is not the case. The whiteboard >maintains necessary state for all nodes in the network no matter how >you spin it. If that state does not exist, is not maintained properly, >or cannot be reached by the client node, DAD using the whiteboard will >fail. At the very least, I think we should drop the word "stateless" >everywhere in the draft. I think you're very right, that's an excellent point. Stateful in the DHCP acceptance is certainly closer to what we are doing. Pascal _______________________________________________ 6lowpan mailing list [email protected] https://www.ietf.org/mailman/listinfo/6lowpan
