I'm interested in exploring that model using FLOSS on Linux:
- CARLA to host the LV2 / SFZ / SF2 / VST instruments on each worker, possibly including arbitrary effects plugins.
- zita-j2n on each worker, routing rendered audio over the network to the conductor machine.
- zita-n2j on the conductor, capturing rendered audio from the workers and making it available as "readable" JACK audio ports.
- qmidinet on both the conductor machine and on each worker (do I need one instance on the conductor for each worker?
- Some way of querying the set of "bindings" (instrument, MIDI input port, j2n/n2j output port) available on each worker.
- Some way of rendering / previeiwing the GUI for each instrument (or at least its automatable parameters and their assigned MIDI CCs).
- Some way of rendering the note-range for each instrument (including keyswitches, etc.)
There are likely limitations I will trip over, e.g. qmidinet appears to allow a maximum of 32 ports. I don't really need the conductor to be listening for UDP MIDI, and the workers don't need to send it, and the UDP broadcast bits seem unnecessary.