On Thu, Dec 14, 2000 at 10:50:33AM -0500, Victor Iannello wrote:
> Paolo,
>
> Paolo,
>
> You are confirming what we suspected--the variation in time required to
> access the timer is a symptom rather than the cause of the jitter. You
> mention three possible causes at the hardware level--bus contention, DMA
> steals, and cache disruption. We have done some experiments with the cache
> turned off that have not improved jitter. So, we are probably at the
> hardware limits of x86 board architecture. It never was designed for
> cycle-by-cycle determinism.
You might want to try, for example, the AMD SC520 board. The combination
of board design and chip design makes all the difference sub 20us.
>
> Regards,
> Victor Iannello
> Synchrony Inc.
>
> ----- Original Message -----
> From: "Paolo Mantegazza" <[EMAIL PROTECTED]>
> To: "Victor Iannello" <[EMAIL PROTECTED]>; "rtlinux mailing
> list" <[EMAIL PROTECTED]>
> Sent: Thursday, December 14, 2000 8:14 AM
> Subject: Re: [rtl] Interrupt Jitter
>
>
> > Victor Iannello wrote:
> > >
> > > If I understand correctly, you are attributing a large cause of the
> jitter
> > > to be read and write operations to timer registers. Because a read of
> the
> > > timer register(s) is performed even in periodic mode, there is
> corresponding
> > > jitter.
> >
> > There is no timer reading in periodic mode. You are right when you say
> > that the jitter is depending on system activity. In fact you are causing
> > more interrupts and, even if it is true that Linux interrupt are
> > deferred processing, they must be at least acknowledged. Nonetheless the
> > jitter they cause is not enough to justify the highest peaks.
> >
> > However there is no chance to get the details of what is causing what:
> > bus contention, DMA steals, cache disruption all cooperate to the mess.
> >
> > Note that the jitter is almost independet from the CPU clock. I also got
> > the idea that having a timer on the PCI bus does not make any
> > difference, the related interrupt will be signalled by the 8259 PIC
> > anyhow.
> >
> > I always thought that if absolutely much less than 20 us jitter is
> > needed you need something different from a general purpose computer.
> > That is what DSPs were born for.
> >
> > Ciao, Paolo.
>
> -- [rtl] ---
> To unsubscribe:
> echo "unsubscribe rtl" | mail [EMAIL PROTECTED] OR
> echo "unsubscribe rtl <Your_email>" | mail [EMAIL PROTECTED]
> ---
> For more information on Real-Time Linux see:
> http://www.rtlinux.org/rtlinux/
--
---------------------------------------------------------
Victor Yodaiken
Finite State Machine Labs: The RTLinux Company.
www.fsmlabs.com www.rtlinux.com
-- [rtl] ---
To unsubscribe:
echo "unsubscribe rtl" | mail [EMAIL PROTECTED] OR
echo "unsubscribe rtl <Your_email>" | mail [EMAIL PROTECTED]
---
For more information on Real-Time Linux see:
http://www.rtlinux.org/rtlinux/
