[LAD] AVB not so dead after all

[Len Ovens]

On Sat, 6 Jun 2015, Reuben Martin wrote:

> I thought I would post this since there was a big conversation here a while 
> back about AES67 and the slow death of AVB due to lack of support.
>
> Well I was talking with a guy from Meyer Sound who told me that AVB has been 
> resurrected from the dead. Apparently Cisco and other large network hardware 
> vendors were willing to back it as long as it was made more generic to 
> accommodate industrial uses that are also time-sensitive.
>
> So apparently it has been re-branded as “Time-Sensitive Networking” and has a 
> lot more momentum behind it.
>
> http://en.wikipedia.org/wiki/Time-Sensitive_Networking
> http://www.commercialintegrator.com/article/rebranding_avb_4_key_takeaways_from_time_sensitive_networks_conference

Interesting.

Some notes on AoIP and Linux. There are some well funded people/companies 
that use Linux for many things, but much of the development in the audio 
world is with people who have hardware that they can't afford to replace 
and so write drivers for. I think this is part of the reason we are not 
seeing much in the way of Linux drivers for AoIP (AVB, AES67, Ravena, 
whatever). Right now, AoIP on Linux costs about twice as much as a normal 
audio card because the Linux box requires both an interface card in the 
computer as well as the Audio IF on the other end of the network cable 
(not to mention a switch in the middle).

Why is this? Linux is based on lowest common denominator hardware... we 
call it the PC. The Linux world has gotten much better preformance out of 
this box than it was designed for. But, in the case of audio, the HW does 
limit performance at least with AoIP. That limit is the clock. The PC does 
not have a HW PTP clock built in and in this case software is not good 
enough. The way around this is with a custom NIC that does. For some 
reason even though one can buy an ethernet chip that includes a stable PTP 
clock for less than $5, any NICs I have found with a PTP clock are closer 
to $1k.

I was "listening in" on a IRC conversation about the differences between 
ALSA and Core audio and why Core audio "does it right". The difference 
ends up being this HW clock. That is ALSA is build the way it is because 
the PC requires it to be.

Whats the point of all this? TSN sounds good to me. It widens the scope of 
low latency networking and the requirement of distributed clocking into 
areas where cost matters. I am hopeful that this means the cost of a NIC 
with good HW clock will go down or even become standard. All kinds of AoIP 
would see the benefit from this. I also think the cost of AoIP audio 
interfaces would come down to similar cost as USB or firewire.

There is no reason we could not make an ALSA AES67 driver that would work 
with any GB-NIC out there but the closed drivers now available show that 
on a PC latency is double that of Core audio and handles fewer channels.
(Core audio at 192K = 64 channels in and out min latency 32 samples, Win 
at 192k = 16 channels in and out min 64 samples) So any ALSA driver would 
suffer from similar lower performance. This is why almost all AoIP setups 
suggest their PCI(e) card in place of your stock NIC.

* numbers from:
http://www.merging.com/products/networked-audio/for-3rd-party-daw
I have seen similar numbers (or worse) elsewhere.

* I am not in any way suggesting anyone use 192k sample rate for audio 
recording or streams. It's use here is only to show the difference in HW 
capabilities. 48k is what I use and suggest others use.

--
Len Ovens
www.ovenwerks.net