[LAD] AoIP question

Len Ovens len at ovenwerks.net
Wed Oct 8 02:31:41 UTC 2014

On Mon, 6 Oct 2014, Reuben Martin wrote:

> On Monday, October 06, 2014 04:10:59 PM you wrote:
>> On Mon, 6 Oct 2014, Reuben Martin wrote:
>>> IGMP Snooping built into the switch is very convenient. Otherwise the
>>> multicast audio transport will flood your ports.
>> IGMP is one of those things mentioned in the AES67 spec. that I have to
>> read still. For some reason they have not bothered to spell out what IGMP
>> stands for as they have with just about everything else.
> IGMP is the protocol used to solicit membership to a multicast group. IGMP
> snooping is the switch probing layer3 in order to monitor which ports have
> connections subscribed to multicast groups, and uses that to determine which
> multicast traffic should be sent to which port. Without IGMP snooping, the
> multicast traffic is sent to all ports, which makes it no different than
> broadcast traffic. And with multicast flooding all ports, you actually eat up
> more bandwidth than if you sent duplicate unicast traffic to multiple locations.

Sometimes, it is a good idea to read about the parts of things one thinks 
they already know. IGMP is a very basic part of multicasting and while I 
understood the idea of multicasting, the Linux multicasting how-to was 
quite a good starting point. It is quite old (pre 2000 and 2.0 kernel) has 
a few mistakes, but explains the basics pretty good. It talks about IGMP 
v2. but I notice this machine (kernel 3.13) has IGMP v3.

AES67, however, requires IGMP v2 and notes:
  IGMPv2 support is required because IGMPv3 devices operating on an IGMPv2 
network experience a two minute startup delay looking for IGMPv3 services 
on the network.

I don't know if the linux IGMP can be set to run v2ish or not (or if it 
would even be a good idea).

With regard to the idea of generating a word clock from ptpd system 
clock... I think this would be pushing the low latency limits of most 
systems. Using this clock to generate a period clock would make more 
sense. I do not know that period clock would be useful for generating a 
word clock in HW though. I can understand why switches with a PTP clock 
are expensive/rare. I can also see why digital audio devices often have 
more than one processor (each having multi cores).

Len Ovens

More information about the Linux-audio-dev mailing list