[c-nsp] Multicast Issue

[Kristian Larsson]

On Thu, Apr 27, 2006 at 08:35:30AM -0600, John Neiberger wrote:
> >In any case, the tools to use now would be sh int cou on gig 4/16 to 
> >see if mcast out is incrementing on the 6k (hopefully it's a fairly 
> >high rate source, if not, then you won't be able to tell), and if 
> >mcast in is incrementing on the 2950.
> >
> >You can also always try a span session to a sniffer of gig 4/16 in 
> >the tx direction to see if the packets are going out.
> 
> Unfortunately, this application sends a single multicast packet. It
> will send up to six if there is no response. This is just a horribly
> designed application. Get this:
> 
> The client begins talking to the server via TCP, so it already
> obviously knows the server's IP address. After it's been talking to the
> server for a few seconds (a couple hundred packets) the client sends a
> single multicast packet to 224.0.1.55. This is some sort of server
> location mechanism. But it shouldn't be necessary because the client
> obviously already knows the IP address of the server. To make this even
> more colossally dumb, the IP address of the server is IN THE DATA
> PORTION OF THE MULTICAST PACKET! If the client does not receive a
> unicast response to this multicast packet, the client can't connect, at
> least from the user's perspective. Really bad design because that means
> that the server doesn't use IGMP to join the group, which means it has
> to be on the same LAN as all of the clients. I'm just trying to find a
> way to make it work for now until we can talk to the developers of the
> application.
You could prolly write your own application that
will send an IGMP join for that group.

  Kristian.