ubuntu10.04 Lcnc 2.7 rtai error

correspond to ICH7_1 ??

No, you have ICH7_2, the _N bit is the rev number

00:1c.0 PCI bridge: Intel Corporation N10/ICH 7 Family PCI Express Port 1 (rev 02)

In my experience the Atom MBs from D945GCLF through the D2800s have quite bad latency
Their latency test results are generally good but if you physically measure
latency by say toggling a parallel port pin you will find that they have 50 to 75 usec latency
_when they actually do something_ The latency test _doesn't do anything_ so only tests
dispatch latency.

Dispatch latency very optimistic compared to actual latency when memory allocation and physical I/O need to be done

If you want an education in whats going on, try plotting (with halscope) some of the hal function times
(watch the plot as you open firefox or move the mouse)


If you get a RTAI error on a 1 KHz thread that means you have at least a 200 usec variation from thread to thread
(The warning happens when the error is more than 20% of the thread time IICRC)


I dont have and real time errors on my D510 running an identical setup so something must be different
The D2550 does have different video than the D510 which might be an issue
(and uses the poorly supported Power VR hardware)

@ ArkEyes ... than not have SMI problem?

... actually running latency test on these pc from 4 hour.... (videos and other ...) 20000 1msec - 23000 25usec

@PCW I'm no expert but I was pretty sure of the real world measurement ( with the parallel port to work for example ) was different that the measure only software ... I use the integrated video interface ....

If you want an education in whats going on, try plotting (with halscope) some of the hal function times (watch the plot as you open firefox or move the mouse)

Sorry I just do it .... I grab hm2_5i25.encoder.05.input-a 4000 sample at 1Khz trigger auto .... if there was a problem, what should I do ? I see only the signal that stay at 1 if don't move the jog-wheel ....

@ ArkEyes ... than not have SMI problem?

No, only the rev 1 chipset is noted to have SMI problem

... actually running latency test on these pc from 4 hour.... (videos and other ...) 20000 1msec - 23000 25usec

The figure for the 25us thread is irrelevant, because you will not be using a base-thread, 20K is perfectly acceptable for the servo-thread.

If you really want to scramble your brains run the latency test with arguments

latency-test 1ms
will run a latency test with just a servo thread, which is what you will actually have.

I have found in testing, that this often produces far worse results, especially on rt-preempt kernels for some reason, than running a latency test with 2 threads, even if one will not be used.

So much so in fact, that I am now running my mill with a dummy base-thread, which is not connected to anything, even though I have a 5i25 / 7i76 combo and only use the servo-thread.
It certainly produces no problems and I am trying to compare that config to a standard one, via an inline latency test I have written, to see if it is actually beneficial in real world machines.
Lot more testing to do.

regards

I believe that you are seeing a caching issue

My experience with Preempt-RT and our Ethernet cards show the same thing,
the faster you set the servo thread, the faster you _can_ set the servo thread
because the LRU cache algorithm keeps the main RT parts cached if they are
accessed frequently enough

I have also noticed that CPUs with larger caches have better RTAI and Preempt-RT performance
Here's a fast CPU with a 3M cache:

Preempt-RT
freeby.mesanet.com/h97-g3258-preemt-rt.png

RTAI
freeby.mesanet.com/g3258-rtai.png

@ ArkEyes .. for experience I try to run latency with arguments ....

@PCW ... than if try the same config with asrock Q1900B-itx (intel 4 core 2M chache) I've the better result ...

My problem at the end si that .... as suggest PCW I try halscope ... noew with 2 channel fo encoder ... but not see nothings eccepts the 2 signal .... 8000 samples at 1Khz .... Wicth is the command to run halscope at 10Khz for example??... I can change by hand the value of jogwheel encoder!

Anyhow ... the realtime issue is show in axis (in touchy too??) and if the error is not a real problem how do you disable the error message so that it is never seen ?

Regards
Giorgio