[ntp:hackers] Reduce jitter of refclocks (connected via USB)
tglassey at earthlink.net
Tue Mar 15 14:31:19 UTC 2011
On 3/15/2011 1:05 AM, David J Taylor wrote:
>>> I once did some casual checks of running NTP with a GPS/PPS source over
>>> a serial to USB converter, and got the results here:
>>> Jitter was in the region of 50 microseconds (Windows XP SP3). Not quite
>>> sure how that tallies with a 1 ms polling interval.
The USB Hub's sampling rate is what causes this.
>> As you write on that page, "However, it may also be that the jitter
>> reduction was due to the shorter polling interval used for the
>> ref-clock than for LAN operation" heavy oversampling will tend to
>> reduce the jitter.
> But I was seeing 50 microseconds jitter, not 1 millisecond.
>> Anyway, it seems like the achievable jitter figure depends a lot on
>> the specific Windows/driver version combo, I'm getting 1-2 ms jitter
>> from the Sure GPS board connected to a real COM1 port on W7-64. :-(
> One millisecond is about the jitter limit for those Windows systems
> which don't use the interpolation code, i.e. Windows Vista and
> Windows-7, at least when not using a refclock.
>> Using the USB connection is _much_ worse though, since that GPS board
>> doesn't even try to send the PPS signal over the USB port, so I end
>> up with 10-15 ms jitter from the NMEA messages.
That's the specific interfaces fault...
> Thanks for your comments, Terje.
> Using a GPS/PPS signal connected to a real COM port on Windows-7
> 32-bit, I am seeing between 22 and 30 microseconds averaged jitter,
> using Dave Harts kernel-mode serial-PPS driver - but it wouldn't be a
> lot worse with the standard driver. I can't use that in a 64-bit
> system as it's not signed, but I may be able to test later in the week
> with my second Sure board. Here are the offset graphs:
> Upgrading to SP1 seems to have helped on that PC (upgrade on 2011 Mar 01)
> I would expect that using NMEA alone would produce far worse jitter -
> I recall that Internet sync produced better results than just serial
> data. I've also seen the "failed to retain baud rate" you noted, so I
> will also see whether I can get it working as expected at the default
> 9600 baud.
More information about the hackers