From: Marc M. <ma...@co...> - 2002-06-25 01:07:29
|
Erik Nordstr=F6m wrote: > On Mon, 2002-06-24 at 22:20, Marc Mosko wrote: > > Overall, I believe you put to much effort in finding loopholes in the > wording, instead of trying to understand what the draft really says. > Don't take me wrong, there are lots of things in the draft that many > would like to see changed. But you must also understand that any chang= es > you propose, someone else may find hard to understand. Let me try to start over. I'm not trying to be difficult about the wording. The issue with 'active' is that it's an important word in the draft so I think it would be good to use it carefully, but is somewhat beside the main issue I was trying to raise. I did a bit too much speculation about possible interpretations of the rules in the previous email. So, please let me rephrase. I will also try not to speculate about what should happen. My purpose in posting this is to get feedback from others about what they think should happen. Main point: the behavior of a node is ill-defined when a 1-hop neighbor becomes a multi-hop neighbor through RREQ or RREP updates to the routing table. The reclassification as a multi-hop neighbor might or might not be true in fact, but it is the nodes 'best knowledge' of the network at a given point in time. Such updates may happen without any data packets flowing at that point in time. I have found in simulation that this may lead to the issues below, which are interlinked. The way one answers them may have noticable effects on performance and maybe interoperability. If someone will say that there's no interoperability problems however you answer the questions, that's great. In specific: (1) If a one-hop neighbor N becomes a multi-hop destination, what do you do to the other destinations that used N as the next hop? It appears by the draft that you do nothing. (2) What should you do with the Hello timer? Since this is not a 'link break' event, it appears that by the rules it will remain running. If it remains running, what do you do when the timer expires? (3) If the next hop N for a route R is not listed in the routing table as a 1-hop neighbor because N was redefined as a multihop neighbor, should a node respond to RREQ for R? By the rules, since all routes are active it appears that you should send a RREP assuming the sequence numbers work out. (4) Is it acceptable to have the next hop for a route be an address that is not a known 1-hop neighbor? I think this would have implications on the pre-cursor lists. Thank you, Marc Mosko U.C. Santa Cruz |