List Of Problem Hardware

• TOSHIBA 100cs
• "no 845/850 based motherboard was able to get below 200 us worst latency"
  • [see RTAI Mailing list Thread 1]
  • [see RTAI Mailing list Thread 2]

Date: Thu, 10 Jul 2003 15:55:13 +0200 (CEST)
Subject: IDE-HDD accesses disturbs realtime timing (RTNET)

Hi everybody,

my setup is a P4 PC (2.4 GHz) with an IDE HDD, I disabled DMA for the IDE HDD.
On the HDD I use the reiserfs filesystem, no swap partition is in use.

I have an Ethernet (Intel eepro100) adapter plugged in the PC, this is used together with rtnet
(realtime ethernet add on to RTAI).
Kernel: 2.4.20, rtai-24.1.11, rthal-5g, rtnet-0.2.9.

I use the ethernet to transfer data in a 10 millisecond cycle to a remote device.
My remote device is configured to change a voltage on an output pin, the value of this voltage is
transfered within the ethernet message.
When sending two alternating values, the result is a rectangle waveform that can
be measured with an oscilloscope.
It works fine (stable rectangle, nearly no jitter) until an HDD access occurs. This leads to a
terrible jitter (up to 40 milliseconds) on the rectangle signal.

The HDD is used for logging, I cannot disable it for the time the system is running.
It looks like there is a conflict on the PCI bus.
Is this possible?

Can usage of DMA for the IDE improve the behaviour? The realtime driver of the
eepro100 from rtnet itself uses DMA - will there be a conflict?

The reiserfs does regular accesses to the HDD (about every 5 seconds) can this be stopped
somehow?

Any ideas on this problem are highly welcome!

Thanks for any help

Mathias

USB

> For any who may be interested, I found the culprit.
> I started to disable all IO peripherals in BIOS that I was
> not using and found that the On Chip USB and USB Legacy Support were the
> cause.  Once these were disabled, I get 15usec max latency.

I have seen this problem with other RTOS systems on other SBCs.  It is usually the USB Legacy Support that causes the problem. Periodic NMIs of 300 microseconds!  Eeek!.