Ghostcast using Mulitcast all in same VLAN

[Migration User]

I'm having some issues trying to get this to work. Let me paint the network picture.

Core switch trunking out to 5 switches.
Ghostcast server & file server on same vlan.

Switches in Vlan 3, Multicast works but really it's broadcasting instead of multicast because the switches don't support IGMP Snooping.

So now we got new switches... that support everything. Core Cisco 6513, Bench switches Cisco 4500's

same type of setup... Although this time, IGMP snooping is turned on for the 4500's, and pim is setup on the 6513. Cisco Multicast looks to be setup correctly. But Ghostcasting is not working. The clients join the Ghostcast session; I see them join on the Ghostcast console. But the clients just sit there and don't image.

But if I turn off IGMP Snooping on the bench switches they start to image fine, but I know they are really just broadcasting. So what I’m looking for is, how to do Ghostcasting all on the same vlan? Without broadcasting all the traffic.

Anyone have any ideas...

[Migration User]

The most important single thing to check - the cause of almost all multicast problems, in fact - is that your router is actually configured to generate IGMP queries. What you describe are typical symptoms for not having an IGMP querier.

If IGMP snooping is enabled in the switches, they see the initial IGMP group subscription sent by the hosts and assigns the switch port to receive traffic from that group. However, that membership will rapidly time out unless it is refreshed, and the IGMP protocol spec requires that a router issue IGMP queries which get hosts to re-subscribe. This process of the router issuing queries and the hosts re-subscribing is what keeps the group memberships live in the switch. If this is not configured, multicasting appears to "work" briefly but then stops.

IGMP querying is independent of PIM, which is a router-router protocol for creating wide-scale groups. You also mention "Cisco multicast" which is presumably CGMP, but CGMP is a router-switch protocol and it does not function as a substitute for IGMP.

The easiest way to verify that at least once device is configured as an IGMP querier is to perform a packet trace using a tool like WireShark. You should expect to see IGMP query messages from the router to 224.0.0.1 (the all-hosts group) fairly regularly, at a frequency high enough for the switches to not time out the host group memberships. A typical querier will transmit every minute, while a typical switch timeout is 3 minutes.

Beyond that, the only thing of consequence related to VLANs and multicast is that many switches have a limit on the number of supported simultaneous multicast groups; when partitioning the switch ports into VLANs, for many if not most switches that internal table ends up being divided amongst the VLANs (so, a switch that supported 32 multicast groups would only support 8 groups per VLAN if there were 4 VLANs). Configuring large numbers of VLANs can thus end up meaning no individual VLAN can support enough groups to actually function using multicast IP.

I don't think that concern is likely to be related to your situation, but if your switches do have a group limit like this, it would be a good idea to find out what it is.

[Migration User]

Thanks for the response,

I checked into what you were talking about and used a sniffer, and I do see IGMP queries happening every minute.  They were coming from the Core switch 10.10.192.30. Just as a test I changed it to every 30 seconds. On the sniffer it started reporting them every 30 Seconds. So I went and tried another test.

On the Ghost console I saw the two clients join the cast and it shows “casting in progress” but if I go look at the client systems their just sitting there not doing anything.  Not sure.. I think multicast is setup right but still no luck. Is there anything I would need to setup on the ghost console, like multicast scope? Or TTL?

[Migration User]

Update:  Just as a test I figured I would try this.

My core is 10.10.192.30, which has my Ghostcast server on it.

My bench_switch 10.10.192.32,  2 client systems

I decided to move my clients to the core switch just to see if the Ghostcasting session would kick off. Which it did. But then it looks like it stopped right around 20% in the process.

Then I moved the Server & Clients to the bench switch & tried again. But this time it did not kick off at all.

Ideas?

Really it's a simply config on the bench swithes.  Interface Vlan 2 10.10.192.32. Spanning-tree portfast on all ports but the trunk. 10Gig fiber trunk back to the core. Using VTP from the core to the bench switch. IGMP snooping is turned by default on the 4500's. I'm seeing the IGMP queries from the core. I know Vlan 2 is working because DHCP & PXE boot is coming from the same server on the core. GhostCast server seens the client connecting then fails.

But still no go when it comes to Mulitcasting.

[Migration User]

Sounds like you've checked the basics pretty thoroughly, it's a puzzler.

I can't explain the discrepancy between the two switching environments, but stopping at 20% is the classic behaviour of IGMP querier problems. Since you're seeing the queries it shouldn't be a problem, but what specific version of IGMP is the querier using? I'd guess that it'd generate IGMPv2 by default and the switches would snoop that fine, but since we'll have to be thorough it's something to check.

Beyond that, probably the main thing I'd do is try and sniff both the server and client ends of one of these interchanges and correlate what's being sent by the server and received by the client to see if there is any difference. Even when the session starts to get into an error state the server side should be sending retries - the resend timer backs off pretty rapidly, but it should become apparent comparing the two ends if anything is getting through at all.

By the way, we do internally have a WireShark plugin for making more sense of the sequence numbers in the multicast protocol we use, and although it's not much use for customers 99% of the time it's useful for just this kind of correlation exercise between traces taken at separate points. If you e-mail me at nigel dot bree at gmail dot com I should be able to send you a link tomorrow.

[Migration User]

Sorry for the late response, I'm planning on doing some more testing next week and I will keep you posted.

Thanks again

[Migration User]

Sorry it's been awhile since my last post, but I was trying to get some extra help from someone at my work who is a little high up then me. He has his CCIE in switch & routing. So anyway.. we decided to try and two Vlan setup.  So this is the current setup...

7609 trunked to a 4506 switch

Vlan 2 = Servers(Ghost cast server)
Vlan 3 = Bench's / Clients
7609 is configured for multicast routing
4506 is a layer 2 switch
7609 has a vlan for the Ghostcast server and a vlan for users
4506 all ports are in the user vlan
7609 is configured for pim sparse-dense-mode on both vlans and also has ip igmp snooping querier enabled

User's can get DHCP, PXE boot and do normal udp pulls from the ghostcast server, just not multicast.

So we spend a few days trying to get Ghostcasting to work, but still nothing. The only way it would start was to remove IGMP Snooping on the 4506's. Here is his email to me, regarding the setup.

"Multicast appeared to be running correctly on the switches.  I did multiple testing with multicast and verified that the switches were seeing the multicast streams.  What seemed weird was the application.  It seemed like the three devices (server and PCs) were not talking to each other using multicast. "

Any ideas?

[Migration User]

Anyone?