Multicasting between 3400cl and 3500yl

[James Nord]

Hi,

I have 2 switches, a 3500yl and a 3400cl.

The switches are used extensivly for IP Multicast and with sources and sinks being located on both switches.

The switches have 6 connections between them - one for each VLAN that we want to replicate between the switches.

The problem I am having is that if a source is present on the 3400 and there are no sinks on that same switch then if a sink joins on the 3500 then the sink will not see the data from the source on the 3400 (if the 3400 has stopped acting as a querier due to the presence of the querier on the 3500). Likewise this works the other way around if the the 3500 sees the 3400 first.

If I set the ports that are connected between each switches to forward all multicast this is exactly what we get - all multicast. This is not suitable as can have more traffic than the link can cope with.

So I want to each switch to prune the multicast - but in order for this to work the active querier would have to forward all joins/group membership reports to the other switch. Unfortunatly this is not happening as the switch blocks all multicast traffic that has not been subscribed and not just multicast traffic of type UDP.

Is there any way around this - or to get the functionality that I am after?

[Matt Hobbs]

Just to clarify, what you want the switches to do is to keep all their multicast traffic to themselves until it is specifically required on the other switch?

What I believe is happening at the moment is that when you connect the two switches together one is becoming the querier for that VLAN which is expected, what then happens is the non-querier will forward all mulitcast traffic to the querier regardless of if there has been a join or not. This is normal, but not what you would like... is that correct?

If I'm on the right path, what you may need is PIM-Sparse which is available on the 3500's with the Premium Software license.

(By the way, you may want to consider consider using a 6 port trunk tagged on all the VLANs instead of using individual links untagged in each VLAN - or even better the 10GB modules)

[James Nord]

You are correct - I want the switches to keep their multicast trafic to themeselves until there is an interested party on the other switch.

You are nearly right in what is happening but not quite.

The switch which is no longer the querier will _not_ forward multicast to the interested parties on the other switch. It doesn't see the membership reports from the parties on the switch with the querier active so stops forwarding the multicast.

I have to some extent worked around this by setting the ports on each switch that connect to each other to "Forward". However this forwards all multicast and not just the multicast that the other switch is interested in.

(I haven't tried using a 6 port trunk as I never had much faith in this on previous cisco equiptment - the 10GB modules are still a tad expensive :-)).

[Matt Hobbs]

That's a bit strange. In my experience if I connect up two switches and enable IGMP, after about 6-7 minutes going through the slow querier election process they will be ready for proper IGMP action.

On one, 'show ip igmp' will list itself as the querier, and on the other 'show ip igmp' will point to the IP address of the querier.

At that point, all multicast traffic should be forwarded to the querier. My understanding is that it occurs this way as the querier needs to know about all multicast traffic, so that when it receives a join it can create a new bridge group and start forwarding the traffic as required.

When a switch receives a Join it only forwards the Join to the port that the querier is on. In your case you would like it so that the Querier relayed this Join on to the next switch, unfortunately I don't think this is possible unless using something more efficient like PIM-Sparse. That way each switch will be a Querier for it's own VLANs and the interconnecting VLAN will only forward on requests as needed.

Once again, this is my understanding only!

Also you shouldn't need to explicitly use the 'forward' option, I would leave the default which is all ports 'ip igmp auto'. Going to the querier port though both are effectively the same in my opinion.