[ardour-dev] Re: [ardour-users] MTC & MMC - CORRECTION

[Paul Davis]

>That would be RTAI (http://www.rtai.org).  Not a particularly fun thing to

No. This is a misconception. The problem in Linux (and I would imagine
Linux and perhaps OS X as well) is that there is no interrupt source
with adequate resolution. The problems related to scheduling latency
and jitter are obviously relevant, but secondary given the complete
absence of a source that can be used to schedule the delivery of N
MIDI bytes every so often.

The way to do this on most modern h/w is to keep reprogramming the
APIC or its equivalent - it has a very, very high resolution crystal
and can be used to schedule interrupts at intervals measured in the
10s of usecs. 

The KURT patches did just that, and now the HRT (High Resolution
Timers) do the same thing. They both toss out the old "system timer"
and instead continually reprogram the APIC so that it will interrupt
whenever it is next needed (this obviously defaults to the system
timer interval if nothing else requires it).

>Of course, most modern hardware from Intel and AMD can do much better than
>that.  I saw reports on the RTAI mailing list of getting timers at several
>thousand Hz working reliably with low-latency and jitter on modern hardware.

2.6 with Ingo's patches appears to work more or less as well as
RTAI. not suprising, since he's essentially provided a mechanism for
virtualizing most of the OS that we don't care about :)

--p