issue.Minutes - Last Mile BOF - Keyur Patel and Greg Shepard MLM - a misnomer Greg - In the enterprise, multicast is easy to deploy, Not so in the global Internet, Enabling multicast one network at a time brings you nothing. Multicast is all or nothing - either it works or you get nothing. Goals - provide a multicast only transport solution for content owners - independent of multicast penetration - leverage existing deployment Toerless E. - Thinks that there is no business demand for tunnels Dae Young Kim - There are two businesses in Korea (one for massive many to many gaming) which use a box for autotunneling if multicast is not available. I implemented a common control protocol to locate and do autotunneling. The first company is called Media Delivery - video on demand like application. Toerless - Is this application layer multicast ? Answer - the server and receiver look only at the multicast addresses. We want to use any native multicasting available. Mark Handley - I think what Toerless was trying to say was that there is a continuum between static IP based tunnels and automatic tunnel on a per group per source basis. Lenny G. Regarding demand - A few months back probably the largest content provider came out with a proprietary box solution that does the tunneling, except it was unicast to unicast. Greg Shepard - After my NANOG talk two years ago about this, I had major content providers come up to me and say "solve this and we are interested" and so I think demand is there. Toerless - I want a service provider to say that they could accept boxes that would provide thousands of tunnels automatically. If you cannot reconstitute an IP multicast packet you do not have a network layer tunnel. Mark Handley - I think that Toerless does have a point here. If there is an application that is producing packets with the same source and the same port, but different groups, then you need the group address to make things work. Mark Handley - we seem to be confusing how do you set up tunnels versus how you use tunnels to emulate ASM with SSM or something similar. Kevin Almeroth - Is there a consensus that this (tunneling solutions) is something that should be done. Michael - I tried for 2 IETF's to get a multicast for Yokohama without success. I think that this should have been done in 1999. Mark Handley - There needs to be a business model to deploy this and we need to be careful not to set ourselves up for denial of service attacks. Lenny G. If you could set this up so that people who have not turned things on can get multicast, then this would create measurable demand. Jon Croncroft (?) We should not be doing business model stuff in an engineering meeting. MBoneD addresses relatively well the wide area deployment, but there are a lot of holes in the last mile - IGMP snooping for example - that could be documented. Michael - We have also had deployment issues with IPv6. Now there is a web site for IPv6 tunnels. This just required some smart web designs. Then we set up 6to4 which requires no cooperation from anyone else. Is there a technology or an idea like that for multicast. Greg - I agree that the 6to4 transparent solution is what I would like to see Toerless - I think that the interesting case is where one tunnel end-point is a host. Bill Fenner - NAT's are also an issue. We need some mechanism to get across NATs which are common on home networks. Lucy as Tom Pusateri - We should not change the signaling protocol if at all possible. Do not make encapsulation a requirement. Do use the router alert bit so that routers in route can trap the joins and send the data if they have them. Dino F. Why do we not start with UMTP and see if it will work. Greg Shepard - There would need to be tweaks to it, but it's close and there is code. Bill Fenner - If you look at Live.com multikit tunneling it resources the packets on the wire using a loopback address so the group address is fine but the source is changed. Dae Young Kim - video providers with very high quality video cannot afford the servers to source high quality In that way this big ISP talks with a lot of broadcasters and say, why don't you connect directly to us. Mostly, the server breaks down before the service provider breaks down. They do have a business model here and are very much keen on this . The homes that have large bandwidth want video. Audience member - This gets around the chicken and egg deployment problems. Some of the edge devices, DSLAM's for example, do not do a good job of supporting mullticast. I think that this is a very good idea and see a good case for this. But, why do transit tunnels not work ? Greg Shepard - Tunnels are one to one. You may think that you are getting a network but you may not be getting much more than one interface on some router in their lab. Dino F. The long term goal is to push multicast out to the edges. In unicast you really don't care and you can do host to host or router to host or whatever. In multicast you have to be a lot more careful. Joelja - Our problems are all at the edge. All of our router interfaces are multicast enabled. The problems are in switches and hosts. Eubanks - suggested the OpenBrick / OpenTv model for a simple open source network appliance for setting up tunnels. Dae Young Kim - the box sits at the layer 2, sets up the tunnel and puts the multicast onto the local LAN - this box only uses the bandwidth for two streams. Radia Pearlman - I don't know why everyone is so focused on a daemon working without privileges. Is it so difficult to get system administrators to install an application. Lenny G. I think that the descriptions of this service as a strong arm tactic to beat tier 2's with a clue stick is right on. We need to do this. We asked nicely and nothing happened. Michael - The whole world doesn't have fiber going through their yard. The copper is mostly used by regulated providers, and the regulators do not care about customer demand. Audience member # 2 - The response of the regulatory agencies is pure speculation. Jon Crowcroft - The content providers (as in RMT) have developed boxes to do application layer multicast. These are in places that would be natural for tunnel end-points, so there may be some commonality of interests here. Greg - Bill, can you offer us any guidance. Bill Fenner - You have a mailing list - you need to send out a summary and see if you can get something happening on the mailing list. Toerless - Would this be in MBoneD or somewhere else ? Bill Fenner - Work can be done in existing places or in new places. We don't know what the requirements and solutions are so I can't say where it should be. |