Experimental Physics and
Industrial Control System

Andrew Johnson <[email protected]> · Wed, 05 Feb 2003 15:53:14 -0600

Dylan Maher Loughnan wrote:
 I am experimenting with EPICS base R3.14.1. running it on a
standard RedHat7.3 machine. I have written device support for a few NI DAQ
cards, which is going very well.

Great. If you're willing to make this code available to other sites when
it's finished, please submit entries for those cards to the EPICS Hardware
Support list. </commercial>
The scan period seems to be consistently extended by an 100th
of a second, when using device support. I first noticed this when
I toggled a binary out at 10Hz to generate a square wave. I saw the
effect very clearly on the oscilloscope.

Here is the ca-monitor with empty device support
 DML:Bo0 02/05/03 16:06:27.070044
DML:Bo0 02/05/03 16:06:27.280062
DML:Bo0 02/05/03 16:06:27.490039
DML:Bo0 02/05/03 16:06:27.700034
DML:Bo0 02/05/03 16:06:27.910037
DML:Bo0 02/05/03 16:06:28.120074
DML:Bo0 02/05/03 16:06:28.330045
DML:Bo0 02/05/03 16:06:28.540051
DML:Bo0 02/05/03 16:06:28.750047
DML:Bo0 02/05/03 16:06:28.960049
Using soft records, there is no such problem.
Has anyone else noticed this?

Yes, and I spent some time looking at it many months ago, although I can't
remember the exact details now. IIRC it occurs because the linux kernel
uses a 100Hz clock tick, and doesn't do timer arithmetic to an accuracy
greater than that tick, so depending on how long your scan task takes to
execute, the system call that implements nanosleep() frequently rounds up
twice, adding an extra 0.01 seconds to the delay.
This is definitely not due to an EPICS bug (I checked that at the time),
and I don't think there's any easy way to get around it other than
possibly explore the high resolution timer kernel patches that are
supposed to exist - I've never tried them though.
- Andrew
--
Tongue-twister: Say "Peggy Babcock" 10 times without stumbling...

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System