On 13 Feb 2006, at 19:15, Andrew Rowley wrote: > It is worth noting that so long as the addresses are different for > each room, there should be no problem if all rooms use the same > port numbers. However the problem is that with dynamic allocation, > you cannot be sure that you will get a different address accross > different venue servers. The short-term solution to this is for > each person running a venue server to make sure that the multicast > address range is different from that of any other venue server. > The long-term solution is to look at some sort of venue server > talking where by the venue servers talk (maybe using a fixed > multicast address) and annouce when they are using an address, so > that the other servers don't use it.
That sounds similar to the old sdr model, with SAP [RFC2974], which started the Mbone off. Unfortunately, it's very hard to make it work right: to keep the bandwidth low you need a large interval between the announcements from each venue, but a large interval such as that interacts very poorly with the timers in the multicast routing protocol. The Cisco folks hate it, and it has caused a lot of the complexity and fragility in multicast routing over the years. There have been lots of other protocols along these lines (look up the history of the IETF MALLOC working group, or the current MBONED group), but none has been successful. Not to say it's necessarily a bad idea, but there's a lot of history to learn from... Colin
