Re: [alsa-devel] Re: It's really the Linux DISK subsystem!
Benno Senoner (sbenno@gardena.net)
Thu, 24 Jun 1999 18:00:28 +0200
On Thu, 24 Jun 1999, Jaroslav Kysela wrote:
> On Thu, 24 Jun 1999, Andy Lo A Foe wrote:
>
> > Hello,
> >
> > Just to rehash a small real wrold test I did earlier. Reading an mp3 into
> > disk cache before decoding and playback (using dd if=mp3 of=/dev/null)
> > produces *NO* underruns. Yes, not one sinle underrun on an 8 minute song!
> > And get this, the fragment size/count was set up as 64 and 3 (with a low
> > watermark of 1). This means 1ms latency!!! (of course the thread in charge
> > of feeding the soundcard is real-time).
> >
> > Gives us some hope no :)
>
> Yes, I always have believed that current hardware is able to reach this
> latency.
>
> Jaroslav
>
fully agree,
current hardware is capable of this, and cached files means that the kernel
lock is held for a much shorter interval, because of RAM access instead of disk
access.
But can someone explain me if in Linux, the kernel can be preempted from
RT processes, during a disk I/O , or must the disk I/O complete to
allow a rescheduling of user processes ?
It the latter is true, then I think it's very hard to do low-latency audio on
Linux.
Does anyone know (just for curiousity) how SGIs manages this ?
regards,
Benno.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/