1

I am trying to identify how accurately my clock has been synced using systemd-timesyncd. I ran the following command however there is a lot of information and it isn't abundantly clear what one is the accuracy.

% timedatectl timesync-status
       Server: 209.115.181.106 (2.nixos.pool.ntp.org)
Poll interval: 34min 8s (min: 32s; max 34min 8s)
         Leap: normal
      Version: 4
      Stratum: 2
    Reference: CE6C0083
    Precision: 1us (-25)
Root distance: 17.973ms (max: 5s)
       Offset: +7.343ms
        Delay: 57.966ms
       Jitter: 2.767ms
 Packet count: 41
Kevin Cox
  • 128
  • 6

1 Answers1

0

Those are what NTP statistics look like. See RFC 5905 about the NTP algorithms, or the performance page in the NTP docs.

Approximately, your NTP server is tens of ms away in latency (delay), estimated a bit better than 20 ms offset from the reference clock (root distance). Plus or minus a single digit number of ms (jitter).

That high of latency is pretty obvious this is via the Internet, even before observing the use of the NTP Pool via 2.nixos.pool.ntp.org.

Think about whether this is acceptable to you. Do you need keep systems at, for example, better than 1 ms from each other? Do you need comparisons to a reference clock even when the Internet is down? Options to do better, if desired:

  • Run your own NTP time service on the local network even closer to your nodes.
  • Delegate to the local hypervisor. For example, the Google Cloud metadata service provides NTP, and the Azure (and Hyper-V) platform provides a PTP clock.
John Mahowald
  • 32,050
  • 2
  • 19
  • 34