[ntp:hackers] Solaris serial port jitter
David L. Mills
mills at udel.edu
Fri Oct 22 09:53:54 PDT 2004
I wouldn't be too surprised if a PPS transition emptied the FIFOs, but
on the Ultra 1 the PPS isn't even connected. The PPS is handled by a
driver all its own, but the driver doesn't do anything unless a PPS
wakes it up. If it wasn't the fact that both the Ultra and Blade behaved
very badly before the PPS was configured and are such lambs after, I'd
put it down as a fluke. I checked the loopstats and the good behavior
has continued since I made the change. Still looking.
Brian Utterback wrote:
> This is fascinating. I have been looking into the driver code for this
> problem, and I can tell you that the drivers do nothing different when
> PPS is turned on unless the DCD signal changes. The TIOCSPPS ioctl merely
> sets a bit flag, and this flag is not even looked at unless there is a
> DCD event, in which case hardpps is called. And that is it.
> This suggests that there is a difference in ntpd when PPS is turned on.
> Could it force a higher priority? I would like to track this down, and
> to do that, I would still suggest using Solaris 10. The diagnosibility
> of these types of problems is orders of magnitude easier with the
> dtrace facility available in 10.
> David L. Mills wrote:
>> Apparently, the serial port driver on both Ultra 1 and Blade 1500 are
>> jiggered so that, if you enable the PPSAPI, the jitter goes away,
>> even if the PPS is not actually connected. Away means a couple of
>> hundred microseconds, but that is expected. It was several tens of
>> milliseconds to over a second. With PPS the Ultra 1 (pogo) is
>> nominally 20 microseconds jitter.
>> That's a relief.
>> hackers mailing list
>> hackers at support.ntp.org
More information about the hackers