Experimental Physics and
Industrial Control System

Jens Eden <[email protected]> · Fri, 05 Dec 2008 12:47:09 +0100

Hi Andrew,

> Thanks for the report and patch.  I would be interested to see the output from 
> an "NTPTime_Report 1" command gives on that system, with or without your fix, 
> since that report displays the measured tick rate as well as the nominal one 
> and your report implies the two were significantly different.
This is the output of NTPTime_Report

NTP driver is synchronized with server
Syncronization interval = 60.0 seconds
Last synchronized at 2008-12-05 10:10:28.001558
OS tick rate = 50 Hz (nominal)
Measured tick rate = 50.002 Hz

> I did wonder whether the NTP time provider should correct for problems caused 
> by an incorrect clock tick rate since it is able to detect them, but I was 
> persuaded by Eric Norum not to do so.
The measured tick rate is pretty close to the nominal tick rate and it took
several hours until the ticksToSkip-counter had accumulated enough ticks to
halt the IOC-Time for about 2 seconds. I consider this as a "normal"
incorrect clock tick rate.
Btw. The problem didn't occur, if running the EPICS example application on the
same hardware. I had to use a real-world application (reading Keitley-devices
via GPIB and doing average calculations). But then I was able to reproduce
it on another IOC with similar hardware.
I believe, as long as the jitter of the osdNTPGet-call was big enough to return
a time value every now and then, which didn't need to skip a tick, the IOC-time
was adjusted and the problem did not occur.

Regards
Jens

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System