Comware

I want to use IGMP to prune multicast traffic on the local network.

We use a variety of 'modern' HP switches (2600 upwards to 5300). We don't use VLAN's on the internal network (other than the DEFAULT VLAN on the switches), but we do on the default gateway (5304xl) which routes the traffic. Although I've read the HP docs on IGMP, I'm really looking for recommendation on best practices to implement it here.

I could just turn on IGMP on the switches, but I need to make sure that all ports can still see the traffic if required.

Any thoughts as to the best way to implement this?

Hi Tony,

For most networks I'd do just that, enable IGMP on each switch and walk away.

If you're just a regular organisation that uses multicast traffic from time to time, e.g. ghosting, then just enabling it will be fine.

If you plan on using a lot of multicast traffic, e.g. video survelliance, tv streaming, then you will need to think a little more about the network design and how IGMP works.

When you enable IGMP on ProCurve switches they go into an election state to decide who becomes the IGMP querier. After about 6 minutes, one of the switches will win this election.

If you're going to be a heavy user of IGMP you will want your core switch to become the IGMP querier, as all multicast traffic will be sent to this device. You wouldn't want all your multicast traffic coming into the core switch, and then being sent out to a querier on the edge.

So what you should do is enable IGMP on the core switch first and wait for it to become querier, and then enable all the other switches. To prevent a switch from become querier you can use the 'no ip igmp querier' command which you may want to specify on your edge switches.

Matt

Hi Guys,

This is interesting for me as well, I have lots of hP switches and we use Norton Ghost to image our desktops, I have noticed that occasionally there is a massive increase in traffic when the engineer responsible is using the ghost cast server. Is this because the server is multicasting and i haven't set up IGMP on the switches?

Cheers Matt (again!). Our core switch is a fully loaded 5308xl, so I might try just enabling IGMP on this first (so it becomes a querier). That makes sense.

Would best practise suggest that I also enable IGMP on the gateway (5304)? This routes across 3 different VLAN's. As this switch is IP Forwarding, it shouldn't forward multicast/broadcast packets anyway, but I'm not sure if it has to be aware of IGMP on the local LAN.

Hi Tony,

It's best to enable IGMP on all switches in a given VLAN, including the gateway.

As you've mentioned though, the gateway will not forward multicast traffic across VLANs, unless you configure a multicast routing protocol such as PIM-DM.

The way that the IGMP querier is decided on ProCurve switches is very passive, only in the event that an IGMP router is not present will one of them become querier.

If on your gateway, you enabled PIM for multicast routing between VLANs, it would immediately become Querier for the connected VLANs on that router. If in the event you had two multicast routers on the same VLAN, the router with the lower IP address will be elected Querier.

Getting a little bit off topic here but it helps to plan out your implementation once you fully understand how it all fits together.

To summarise, it's best to enable IGMP on all switches in the VLAN, and in the event of no multicast router being present, try and have your querier in the core to ensure that multicast traffic is handled as efficient as possible.

Johnathon, enabling IGMP on your switches will prevent multicast traffic from flooding out all ports during a multicast ghosting operation. It will only go to the clients that are specifically requesting it.

Matt

Hi, is there an official HP recommendation regarding IGMP configuration? from what I've read here I would

1. enable igmp on the required vlan. I would enable it on all switches that handle igmp traffic :
conf t

vlan XX ip igmp

2. on the core switch I would configure it to be the querier
vlan xx ip igmp querier.

I've read that the querier is responsible for checking of the clients of an multicast group are still active? Or what's the purpose of an querier?  And what if this node goes down? Can I configure a second switch, which would be the next querier?
Thanks very much!

#IGMP

Thanks Matt. I intend to keep the multicast traffic on the local LAN, and have no intention of routing it.

I'll keep the querier role on the network core and setup the other switches afterwards.

As an aside, we finished deploying
multicast recently. Project details
are here:

http://www.freebsd.uwaterloo.ca/twiki/bin/view/Test/MulticastProject