On Sat, Jul 12, 2003 at 12:50:49PM -0700, Mark Knecht wrote: > Daniel, > One sort of urban legend that's out there in the Linux audio > community has to do with the value of a dual vs. single processor > systems and how much they can help in this area.
It depends what you are looking for. If you are worried about dropouts, the main issue is ensuring that the kenrel driver does not get starved for CPU and doesn't allow the playback/record buffer to underrun/overrun. Audio hardware buffers are generally small, so you generally want to ensure that your audio driver is getting CPU regularly. Preempt can help with this. Another issue is the priority your audio device is given on the PCI bus. This can be tweaked for special-purpose situations to prevent garbled audio if you are pushing your PCI bus to the max. You can read more about PCI latency in part 2 of my Linux hardware stability guide: http://www-106.ibm.com/developerworks/library/l-hw2.html Third, in theory, yes, a SMP system can allow a process to run unhindered by other processes on the system if you have things set up right. This can be done using "CPU affinity" features which are now in 2.5 as well as 2.4 kernels with the O(1) scheduler (which most of our kernels have.) The real key is first figuring out what problem you want to solve, and then figuring out what is causing that problem. It can be many things, from high system loads, to "unfairness" (or too much fairness) in the process scheduling, or possibly your PCI bus being overloaded or due to PCI device latencies being set wrong, or something else. I think that a modern uni-processor system with preempt and the low-latency patches enabled would be great for audio work. I don't think you'd specifically need SMP, unless maybe I'm unaware of some special situations where it would be needed. Hope this helps, -- Daniel Robbins Chief Architect, Gentoo Linux http://www.gentoo.org
pgp00000.pgp
Description: PGP signature
