Subject: Re: [linux-audio-dev] non-destructive editing
From: Benno Senoner (sbenno_AT_gardena.net)
Date: la loka 09 1999 - 22:31:00 EDT
On Sat, 09 Oct 1999, David Olofson wrote:
>
> You can measure average seek time and sustained transfer rate individually, and
> then calculate the buffer size as a function of the number of tracks, and the
> data rate/track.
>
> Keep in mind that the shorter the benchmark is, the greater the risk that
> you're measuring the average rather than the worst case. And only the worst
> case is relevant if you need high reliability...
but worst case could be very very bad, if you take into account that the user
could fire up netscape (from the same disk) during your recording session.
:-)
>
> > BTW, I know they were running OS/9 or something like that which seems to be
> > some kind of UNIX-like OS... Can someone provide details on it just for the
> > fun ?
> > (there might be some kind of Ingo's patch for Fairlight OS that must have
> > been
> > devised back in the eighties ;-)
>
> QNX is another example of an OS used for that kind of stuff. They're dedicated
> real time operating systems from the ground up, and don't need patches to get
> that kind of performance.
Ok QNX is a nice realtime OS, but actually I see no advantages by using it in
a highperf multimedia enviroments, since if you get below the 32samples/fragment
mark the inherent CPU overhead increases (OS can't help here) exponentially,
not to mention that QNX has to perform a context switch (because it's
a microkernel ) when you do some syscall() in your short dsp loop,
and full context switches at >1kHz rates HURT A LOT !
:-)
PS: the cygnus embedded linux (RTOS with selectable features at compile time)
seems to be very interesting, since you can leave out the uneeded stuff,
and maybe get even better than QNX since it's not microkerneled.
Linux seems to hurt even commercial RTOS vendors quite a bit these days.
:-)
regards,
Benno.
This archive was generated by hypermail 2b28 : pe maalis 10 2000 - 07:27:13 EST