Subject: Re: [linux-audio-dev] Audio over Ethernet?
From: Anders Torger (torger_AT_ludd.luth.se)
Date: Wed Apr 14 2004 - 20:56:34 EEST
On Wednesday 14 April 2004 18.08, Francois Dechelle wrote:
> IIUC, you don't need some kind of load balancing between the
> machines, i.e. the applications will be assigned statically on the
> different nodes without any dynamic job creation/deletion?
The convolvers on the nodes must be set up once in startup, getting
their filter graph, and their coefficient sets, and then that cannot be
changed in runtime (apart from changing delays, volume switching
coefficients on the filters etc).
In the first implementation I would make this a fully manual process,
that is you would have to set up each convolver manually. But in a
future version it is thinkable to have only one configuration file
which specifies the whole graph, and then the graph is automatically
distributed on the cluster.
BruteFIR, which is the convolver in question, today supports multiple
processors, but its CPU concept could fairly easily be extended to
incorporate any CPU, not just the ones available on the local machine.
BruteFIR does do automatic CPU balancing (which can be manually
overridden), which could be extended to multiple machines.
Apart from transferring massive amounts of audio on the ethernet, also
commands for changing filters and settings must be transferred, and
perfectly synchronised, which is a challenge too.
I really would like to do some experiments on this, the problem is that
I don't have any equipment for doing it... at least not yet. And a
second problem is that I don't really have any use for it myself :-),
since I can do all my convolver needs on one machine.
/Anders
This archive was generated by hypermail 2b28 : Wed Apr 14 2004 - 20:58:51 EEST