[LAU] Linux audio app design: The nature of xruns

Nature of xruns I never paid attention to mentions of xruns since I rarely got them using a stock Fedora 8 kernel (and F6 before that) on a x86_64 architecture, 4GB RAM. Yesterday though, I got plenty of them and my observation is that in this case any real-time feature associated to the kernel does not matter. In this case the xruns are made by the way applications handles stuff (whatever that is). So that makes me think that there ought to be a guideline, a best practices in developing Linux audio/jack applications. Here's the observation setup. First, the song. The song does not matter, It always plays fine. The song is made by the use of Seq24 which drives several soundfonts loaded by Qsynth in different engine instances, as well as two sounds made by Zyn. Qsynth has several instances of FluidR3_GM (120MB) and one SGM (250MB) loaded. Second, the xruns. I take the outputs and drive them through jackmix, add electric guitar line-in from a Vox amp, and have the outputs of jackmix feed into Qarecord. Start recording. The .wav result is polluted with xruns. Third, same song w/o xruns. I terminate jackmix and Qarecord and launch Ardour. I run all outputs into Ardour, and run the electric guitar as well. Add some Freeverb and MultiEQ. Record. No xruns at all in the resulting .wav file. From this it is obvious that the application layer helps a lot in promoting xruns. It is safe to assume that Qarecord is a quick hack. But, there's seemingly a big difference in application design regarding the handling of audio/jack stuff between Qarecord and Ardour since one produces a festival of xruns and the other none. My question is, is there a guide/HOWTO around that addresses the proper design of audio/jack applications or, is - in 2009 - the only to know about that is by actually going through the source code of successful Linux audio/jack applications ? Cheers.