[time-nuts] Re: NIST NTP servers way off for anyone else?

[Steven Sommars]

I suggest plotting one-way delays versus time rather than histograms.
Recent example: I regularly monitor a stratum1/GPS NTP server located in
Canada.
Delay plots from two completely separate monitoring systems are attached
below.  When the two one-way delays change as seen beginning 2021-12-15
07:00,
(request "delay" increases and response "delay" decreases or vice versa)
the most likely causes are server and/or client clock drift.
If the client drifts, the effect will be seen for all monitored servers.
I use multiple monitoring clients to further minimize false alarms.

Changes in network delays (e.g., due to routing) affect the
request/response delay without the correlation seen below.  Network path
changes tend to result in sudden delay jumps.

These are good heuristics, but could fall prey to a sophisticated network
attack.
For the Canada stratum 1 example, the issue is real, poor holdover
performance.  The administrator plans to replace the current indoor puck
with an external antenna.

Other notes:
I was told that  time-e-b.nist.gov received a server hardware upgrade
today.  My monitoring shows dramatic improvement in IPv6 performance.

Steve

ISP=Comcast (cable)
[image: image.png]

Another ISP 300Mbps fiber
[image: image.png]
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image.png
Type: image/png
Size: 60661 bytes
Desc: not available
URL: <http://febo.com/pipermail/time-nuts_lists.febo.com/attachments/20211216/db33d960/attachment.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image.png
Type: image/png
Size: 49061 bytes
Desc: not available
URL: <http://febo.com/pipermail/time-nuts_lists.febo.com/attachments/20211216/db33d960/attachment-0001.png>