1

I have two RPis running on the same network, with the same Internet access. One of them is a B and the other one a B+ model. Both run the same distro with ntp enabled for quite a long time (this is to say that ntp had time to stabilize)

ntpq -p on each of them looks different, though:

B+:

     remote           refid      st t when poll reach   delay   offset  jitter
==============================================================================
+vps.jre655.com  103.1.106.69     2 u 1922 1024  376  430.041  -83.669  31.648
-bertrand.lfbsi. 194.177.34.118   3 u  492 1024  377   60.881  -14.222 100.617
*ntp1.m-online.n 212.18.1.106     2 u  519 1024  377  153.250  -54.473  60.916
+v.bsod.fr       138.195.130.62   3 u  464 1024  377  165.340  -47.092  60.011

B:

     remote           refid      st t when poll reach   delay   offset  jitter
==============================================================================
*server6.webster 193.190.230.65   2 u  653 1024  377   30.511   -0.418   0.477
+chimay.drazzib. 213.251.128.249  2 u  819 1024  377   26.110    0.720   0.651
+mail.nexxea.com 225.7.21.157     3 u  804 1024  377   29.410   -0.917   2.779
-ip-46-105-189-2 181.92.40.112    3 u  863 1024  377   29.524   -1.160   0.684

Why such a difference? I am asking out of curiosity (ntp does its job on keeping the time right)

Clarification: I understand how ntp works and that there are better and worse servers. What surprises me is

  • for a small variation of results on a given RPi,
  • how these results differ between the two RPis
0

Look at the delay column for B+, the configured servers seem to be on the other side of the globe. Are you using 1.COUNTRY-CODE.pool.ntp.org ?

  • Thank you. I clearly overthought that without thinking about the obvious reason. I was using the stock (debian) servers instead of the geographic ones. – WoJ Jan 30 '15 at 8:13
2

Each device will be connected to a random selection of NTP servers to automatically load-balance the overall system.

Otherwise the temptation would be for everyone to connect to the same tier 1 servers.

  • yes, I know that (I will clarify in the question). What I wonder is why the offset/jitter/delay is so different for each of them, with a small variance of servers per RPi (for instance jitter 30-100 on one of them, and 0.4 to 3 on the other) – WoJ Jan 28 '15 at 8:59
  • No idea. Is one of your Pi's connected wired and the other wirelessly? – joan Jan 28 '15 at 9:01
  • no, both are wired – WoJ Jan 28 '15 at 9:01
1

Sounds like a variation on Segal's Law. If you are running at least one R Pi 24/7, then that one could be made to be the ntp server for the other/s. This would also make for a slightly better netizen by less poking of the busy ntp pool.

Also, ntpq is a snapshot, which cannot give a full picture. I monitor my ntp system with NTPviz which produces lots of graphs/analyses from the ntp logs and shows my R Pi to be long term consistent. Strictly, NTPviz only works with the forked version of ntp (NTPsec) but the two are currently (2016) compatible.

0

The "pools" that are set up with the standard stock installs of Raspbian and ubuntu and all Windows computers are many hops away from the Stratum-1 Master Clock.

Often the pools are Stratum-16.

This means that it is a crapshoot as to whether they are getting the correct time. It can vary by quite a bit.

The solution is to use the Stratum-1 Master clock.

Here is a Q&A on this Stack Exchange that explains how to make the change:

How do I set Raspbian to use the primary time server time.nist.gov?

It is quick and easy.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.