[j-nsp] expected multicast forwarding behavior with igmp-snooping and local igmp querier

Phil Mayers p.mayers at imperial.ac.uk
Wed Sep 18 04:30:03 EDT 2013


On 09/18/2013 12:47 AM, Andy Litzinger wrote:

> This does not appear to always be true.  I obviously haven't tested
> every multicast address, but it seems that pretty much all multicast
> traffic directed to 224.0.0.0-239.0.0.255 will cause the switch to
> flood traffic to all ports in the vlan.
>
> But addresses from 239.0.1.0 and up seem to work as I expect.

Sounds like you're running into the fact that >1 multicast group range 
maps to 1 multicast MAC range because there's not enough room to fit the 
28 bits into the multcast mac lower-half.

239.0.0.0/24 maps to the same MAC address range as 224.0.0.0/24 and the 
latter is defined as "no snooping" because it's the link-local/control 
range, hence both flood.

It's a cisco document, but see here:

http://www.cisco.com/en/US/tech/tk828/technologies_white_paper09186a00802d4643.shtml#wp1002391

Some newer equipment (e.g. Cisco Sup2T) does IP-based multcast snooping 
- it reads into the IP header, rather than relying on the multicast mac 
- which solves the problem. No idea if the EX4xxx can do this, but I 
doubt it somehow.

Basically, don't use (224-239).{0,128).0.0/24 for multicast.

> also- what I'm trying to do is relatively simple but maybe I'm going
> about it the wrong way.  I have groups of servers in the vlan that
> use multicast packets as a periodic "heartbeat" to keep track of each
> other.  I'd like to make sure the multicast heartbeat only goes to
> other servers that subscribe to the same multicast address- not send
> it to every server in the vlan.  does my config seem like a valid way
> to do this?  I don't need to route the multicast across subnets.

FWIW we've had big problems with apps that do that. IMHO a 
broadcast-discovery / unicast-keepalive is superior to those kinds of 
multicast solutions. App developers often fail to test in realistic 
environments, and don't account for the myriad of ways that multicast 
can go wrong (e.g. the above issue!). We discourage multicast heartbeats 
for that reason.


More information about the juniper-nsp mailing list