Yeah, it probably wouldn't be technically on-demand, but a 30-sec to 1-min buffer should work okay. I guess I'll see how good the service is when it gets here, and start hacking.

I have quite a bit of programming experience, some of that hackish, but little Linux experience, and no Empeg experience, so I can't really say whether it's feasable or not from the Empeg standpoint, but it's so cool that I hope that "the boys" build in any necessary hooks. :)

3rd-party programs should be able to manipulate playlists and current songs anyway. That seems like a no-brainer.

Re Napster vs. gnutella, I've had little personal success with gnutella, and much with Napster. The interesting thing about Napster is OpenNap, an open-source Napster server:

http://opennap.sourceforge.net/

I already use several OpenNap servers, and I suspect that if Napster runs into difficulty, the number of OpenNap servers will increase dramatically.

Gnutella has some pretty big problems with scalability as currently implemented.

An extra Napster bonus is that it gives you bitrate and length information -- useful for verifying a streamable file, and verifying that it's the "proper" length -- since without a good interface, I suspect that the software will have to be intelligent about which file to pick, switch servers if one is being too slow, etc.

Alex