AW: RTAI calibration question

[Kai Bischop]

> There has been an answer already. I would like to add a few suggestions:
> - use of calibrate is Ok but my favourite way is the use of 
> "showroom/v3.x/kerne/calibration" where you can play inreractively with 
> all the calibrations;
> - for the in kernel latency get a feeling of the differences that exist 
> between using lxrt or ksched;
> - check differences between having a Linux load or not.
>
> That said you'll see that there is no unique figure. For example you 
> have a machine that is almost double the speed of mine, show a better 
> ISA bus access but your latencies are almost doubled. Is it a 
> Celeron/Centrino?

These high latencies bother me most. The machine is a VMEbus system with a
Gefanuc VMIVME 7807 VME single board computer. It has a Pentium M with 1,8
Ghz and a Intel 6300ESB chipset. 

The system is running Debian 3.1. I can boot this machine with two different
kernels/RTAI versions.

The first installation consists of a 2.4.32 kernel, patched with a legacy
RTHAL, and RTAI3.0r5. The second installation uses a 2.4.32 kernel, patched
with the proper IPIPE patch, and RTAI-3.3-test4.

The RTHAL/RTAI3.0r5 gives me the following result when I run calibrate:

SETUP_TIME_8254 	1859
LATENCY_8254	3610 (ksched)

Results from calibrate with IPIPE/RTAI3.3-test4:

SETUP_TIME_8254	1859  (same value as RTAI 3.0r5)
LATENCY_8254	7183   (ksched, nearly doubled with RTAI3.3-test4)

I don´t really understand that, it´s the same hardware, same linuxsystem.

Can you give any advice, why the latency is so bad under RTAI3.3-test4 ?


> 
> Generally speaking for those that are not taking extreme timings, say 
> less that 10/20 KHz I suggest using a calibrations value that give a 
> negative average, thus anticipating sheduling time a bit so that to use 
> the beasy wait scheduler featur to be as sharp as possible on the 
>  deadline.
>
> Paolo.

Regards,

Kai