gerard van dongen
gml at xs4all.nl
Fri Jun 3 15:18:27 CEST 2005
On Fri, 03 Jun 2005 06:34:45 +0200, Matthew Polashek <matt at tinysongs.com>
> Does anyone here know if it's worth my time to investigate clusterin a
> few 64 bit
> AMD Linux machines to make a massive PD box? Does that add latency?
> Any ideas?
It all depends on what you want to do.
I think normal clustering (beowulf-type) would be very difficult since pd
is not written with that kind of distributed processing in mind.
The easiest way would be to run different pd's on the different machines
and have them communicate through OSC or netsend and netreceive.
Here is becomes a matter of what you want to do.
Sample synchronized synthesis distributed across computers routed to a
central computer with the main soundcard is a different problem from lots
of more or less independent "instruments" using their own sound card.
The latter is much easier of course. And easiest of all is one computer
for sound and one computer for graphics talking to each other over the
You could try jack-udp for sample sync stuff. There are a couple programs
in beta stage floating around that try to do jack over ethernet. With
various levels of syncing.
The basic idea is that on the master machine the are regular (or
in-process even) jack-clients that connect to the network and on the slave
machines a network driver running the jack graph, ignoring the local sound
But people have also been working on syncing multiple machines with
It is all very beta software AFAIK, and the people working on it are
testing mostly with just two machines I think.
YMMV as they say, it sounds like a fun way to spend the summer at least
More information about the Pd-list