[Note: the host my incoming email is on is hosed, so I can't currently read email] Multicast Last Mile BOF Report The problem being tackled isn't completely clearly defined, but the general philopshy is that multicast is not being deployed because most potential receivers are using ISPs that don't supply multicast service. The proposal is that we need some form of auto-tunnelling protocol which would allow a receiver to request unicast tunnelling of multgicast traffic from a replicator somewhere on the multicast-capable internet. Such proposals have come up repeatedly over the last ten years. There isn't a single proposal on the table, so there are a lot of different ways this might be done. The chairs seem to have in mind sending some kind of UDP-based join message towards the source, with the router-alert option set. When the packet reaches a replication-capable router, the router joins the multicast group, and tunnels the multicast traffic to the receiver. The proposal seems to be focussed on SSM, and only on tunnelling to receivers, although there was a little pushbak on this. Other people have slightly different solutions, depending on how the tunnelling is done, whether the original soure and multicast addresses are preserved, and how the search for the replicator is done. There are a number of concerns, including security, potential for using this for DoS, and whether there's a business model for the replicator. Much of the discussion ratholed on whether there's a business model, and to what extent this works around existing ISP's business models. There seems to be a reasonable consensus in the room that something should be done. I think the idea isn't inherently bad. The devil's in the details. I know Bill said this wasn't supposed to become a WG, but I suspect the chairs don't feel the same way. There's probably consensus in the room for a WG, although not consensus on a particular solution. I'm not 100% convinced, but it wouldn't be a rediculous idea.
