Re: Audio link two computers in real-time: ethernet?

[van Sinn <vansinn@post.cybercity.dk>]

Rainer Straschill wrote:
> Per said (about using S/PDIF or ADAT to send audio between two 
>computers):
> 
>>Good things is you don't get any extra latency in the signal, since you 
>are not taking that extra AD/DA tour when sending the already digitized 
>signal directly into the other computer's digital input.
> 
> 
> That is not correct. The main part of the latency of two computers
> interfacing are the interfaces' audio buffers (which are usually in
> the ms range) - the latency of the converters are (at least) one order
> of magnitude below that.
> So by using digital interfaces, you're able to reduce your
> inter-computer latency from e.g. 6.3ms to 6.0ms.
> (additional reading: http://www.moinlabs.de/e_lat.htm)
> One reason to do as you describe is that a) you don't do additional
> converter passes, b) you can sync your entire setup to the most stable
> interface clock of all.
> 
> 
>>But I think it has to be the gigabit ether net, not the older and slower 
>version that are still to be found in old computer hardware.
> 
> 
> There's an issue with ethernet (or rather its implementations on
> typical workplace computers), and that is that the computers are not
> able to handle a high-bandwidth ethernet data stream with low latency
> (sic!). They do perfectly well deal with those stream on average,
> however, while a constant throughput of even 100MBit/s (aka the older
> slower version that are still to be found) will create problems.
> 
> Incidentially, Steinberg uses digital audio interfaces (as you
> suggested) when spreading system load over different computers, not
> ethernet.
> 
> Best,
> 
>             Rainer


I agree with the above.  Haven't tried this myself, but if only older 
(100Mbit) ethernet interfaces are available, it's possible (and fairly 
easy) to do ethernet-over-firewire, which might work better.
But as said, haven't tried this for audio, nor will I be able to test it 
in a foreseeable future.

Another possible method, if interfaces aren't capable of handling a 
contineous datastream, might be bundling two interfaces/ethernets, to 
double the bandwidth.
Such bundling is in IT often used to have fault-tolerent connections (if 
one connection goes down, the other will hopefully still work), but can 
also be defined to do equal-sharing of bandwidth, which is what I'm 
talking about.
Modern operating systems either have this feature, or it can be added.
Again, haven't tried this for audio.

Hope I didn't confuse matters with geek talks ;)

-- 
rgds,
van Sinn