On Wed, 2008-01-09 at 10:43 -0500, Raj Jain wrote: > On Jan 9, 2008 9:31 AM, Paul Kyzivat <[EMAIL PROTECTED]> wrote: > > > If its fully distributed with no central point of control, then > > presumably the individual line cards (UAs) become available/unavailable > > independently. That seems inconsistent with the desire to have a single > > registration enable routing to them all. It would seem that each would > > be needing to register independently. > > While correct in theory, the issue with that is explosion of REGISTER > messages. With 'n' AoRs where each has 'm' Contacts, you would have > n*m REGISTER messages at the start up and then at every refresh cycle. > > It's not that SIP doesn't support what we're trying to do, but we're > basically after optimization. Let's review the key question again: > > Does it make sense to use FQDNs in Contact URIs? Is the Contact > fundamentally supposed to identify a device? There may a couple of > corner cases (like the one I stated earlier) where it might be okay to > use FQDNs in Contact. However, what happens when someone sends an FQDN > in the Contact URI in a 200 OK and then ACK goes to wherever the DNS > tells you to send it to (assume non-record-routing proxies in the > middle)?
I would ask this as a different question: Does it make sense to use REGISTER to control routing of entire networks of devices on this scale? and my answer would be 'No'. -- Scott Lawrence tel:+1.781.229.0533;ext=162 or sip:[EMAIL PROTECTED] sipXecs project coordinator - SIPfoundry http://www.sipfoundry.org/sipXecs CTO, Voice Solutions - Bluesocket Inc. http://www.bluesocket.com/ http://www.pingtel.com/ _______________________________________________ Sip-implementors mailing list [email protected] https://lists.cs.columbia.edu/cucslists/listinfo/sip-implementors
