Am Do, den 30.09.2004 schrieb Erik Nordmark um 18:22: > Christian Schild wrote: > > > I would have expected, that immediately when an interface removes its > > (last) global address, it will try to obtain a new one, sending out an > > immediate RS. This is not the case on various platforms. > > [..] > > Shouldn't a client react when its valid lifetime times out and the > > interface loses the address? [..] > But that wasn't the root of your question. > Doesn't your case fall in a misconfiguration of the routers? > For instance, I've tested things in the past with a periodic RA > annoucement of 10 minutes with a valid lifetime of 5 minutes and > observed that the loose the address after 5 minutes and regain it when > hearing an RA 5 minutes later. That's a fine test case, but it wouldn't > be a useful configuration for an operational network. > > So why do you think we need to change something? [..] > You seem to want to solve the problem when the routers have been > configured with a bad set of parameters to use in the advertisements. > Why can't this be fixed by configuring the routers properly? > > If you think the hosts need to do more work to handle misconfigured > routers I fear it is a very slippery slope; for instance, should the > hosts guard against the routers advertising incorrect prefixes as well?
I agree to most of your arguments. The valid lifetime timeouts can only occur when something is wrong. The RFCs handle prefixes and their lifetimes perfectly well. But still, the real world out there is often different from the nice and clean behaviour described in an RFC. In fact, misconfiguration does happen, be it because of ignorance or by mistake. Also, packet loss it not so uncommon in the real world. There were times, when in some areas of our network we had close to 1 percent packet loss, and it was accepted, because it was just to expensive to fix it. Sometimes an admin doesn't even know that there is packet loss in the network. What is new in IPv6 and SLAAC, is that packet loss might lead to a complete loss of connectivity. I admit that the probability is not very high, that a node misses some consecutive RAs, but it is possible. And it is hell for a network admin to find out, why "some" clients loose their connectivity for "some time". What counts in the end, is that the customer gets a non-interrupted service. The problems above can all be solved, when the client is allowed to react itself, when it loses it's connectivity. The list in RFC2461, when it has to send RS, is a good list in what cases this might happen. Timing out of the valid lifetime is just one more of them. So why not give the client the opportunity to heal himself by adding something like "an implementation may choose to send a RS after the valid lifetime times out" to the RFC(s)? So long, Christian -- JOIN - IPv6 reference center Christian Schild A WWU project Westfaelische Wilhelms-Universitaet Muenster http://www.join.uni-muenster.de Zentrum fuer Informationsverarbeitung Team: [EMAIL PROTECTED] Roentgenstrasse 9-13 Priv: [EMAIL PROTECTED] D-48149 Muenster / Germany GPG-/PGP-Key-ID: 6EBFA081 Fon: +49 251 83 31638, fax: +49 251 83 31653 -------------------------------------------------------------------- IETF IPv6 working group mailing list [EMAIL PROTECTED] Administrative Requests: https://www1.ietf.org/mailman/listinfo/ipv6 --------------------------------------------------------------------
