Subject: Re: [linux-audio-dev] can't get any dropouts?
From: Eli Brandt (eli_AT_v.gp.cs.cmu.edu)
Date: to elo 12 1999 - 11:01:00 EDT
rob wrote:
> the reason linux has crappy timing is because the kernel is
> non-reenterant and can hold onto the processor for relatively long periods
> of times servicing disk access, etc (iirc).
Yeah, the IDE drivers are reported to be tools of the devil in this
respect.
> the rt problem on the other hand has already been solved by the
> kids at the rtlinux project.
But RTLinux doesn't do anything for Linux itself -- it's an RT monitor
that happens to run on the same machine as Linux. As you say: no
protection, no system calls, bare-bone synch primitives, and you talk
to the Linux subprocess through the moral equivalent of a serial line.
Sure, people got a lot done under this sort of programming model back
in the Dark Ages, but I'm fond of running under an OS; it helps my
productivity. Personally, I'd gladly trade off some average-case
speed to get Linux's worst case to the point where I could do
millisecond-range RT programming with POSIX RT under Linux proper.
For microsecond-range, yeah, my guess is that RTLinux is the right
way.
> (it would be a huge pain to modify the
> kernel itself to be reenterant which is why they didn't do that).
I bet the Linux-kernel community could muster up resources to match
the efforts of Sun or SGI or IBM, but it _would_ be a grievous pain in
the ass, and the idea would probably... not match Linus' aesthetic.
Rather than go all the way to preemptability, though, we can get some
mileage out of getting rid of too-long critical sections. Uh, "`we'".
--
Eli Brandt | eli+@cs.cmu.edu | http://www.cs.cmu.edu/~eli/
This archive was generated by hypermail 2b28 : pe maalis 10 2000 - 07:25:52 EST