Sam, While I understand the virtue in behave-compatible nats, how realistic is it to believe that any service provider is going to allow a consumer to directly signal their infrastructure? This assumption was the failing of RSVP as an endpoint qos tool. There is lots of noise about ISPs just putting up massive nat farms to push their customers out, but when it comes right down to it there is no way that will work for anything but the most trivial client apps. All of the assumptions about nat working today are built around local control of the mappings. When that mapping function has to move to a third party, all bets are off. Worse, when that third party has strong disincentives which keep them from allowing their customers to punch holes, there is no chance that apps will work. ISPs are disincented by even the simple issue of after-the-fact diagnostics being more complicated by dynamic mappings, let alone the problem of conflict resolution between customers. Behave compatible nats are a nice concept for enterprises with multiple levels of internal nat, but third-party trust issues will kill any real deployment of a signaling based approach. Tony > -----Original Message----- > From: Sam Hartman [mailto:hartmans-ietf@xxxxxxx] > Sent: Wednesday, December 19, 2007 12:19 PM > To: alh-ietf@xxxxxxxx > Cc: 'IETF Chair'; ietf@xxxxxxxx; iaoc@xxxxxxxx; 'John C Klensin'; > dcrocker@xxxxxxxx; 'Pete Resnick'; iesg@xxxxxxxx > Subject: Re: IPv4 Outage Planned for IETF 71 Plenary > > >>>>> "Tony" == Tony Hain <alh-ietf@xxxxxxxx> writes: > > Tony> the right experiment. It is not right because it does > Tony> nothing positive, other than the threat -maybe- spurring > Tony> some action. A more realistic experiment would be to run the > Tony> entire week with a double-nat for IPv4 (and nats between the > Tony> access points to simulate consumer-to-consumer > Tony> configurations), where the most public one has absolutely no > Tony> provision for punching holes (because realistically an ISP > Tony> is not going to punch inbound holes for its customers, or > Tony> allow them to). > > I strongly support this experiment and believe it would be a really > good idea to run. I do think behave-compatible nats should be used, > but besides that, I think the experiment you propose is far more > valuable than the v6-only experiment. _______________________________________________ Ietf@xxxxxxxx https://www1.ietf.org/mailman/listinfo/ietf