The sensor's clock¶
Drift over an hour without network time, 2026-10-23¶
The Pi 5 on the cable to the Mac (linkside-link: IPv6 link-local, no gateway), up since 2026-10-21 12:53 and idle: both A9000s in monitor, no capture. systemd-timesyncd runs, the default of Raspberry Pi OS, but reaches no server: no default route, Packet count: 0 since the boot, System clock synchronized: no. The Mac on NTP. Thirteen linkside-cli clock lines, one every five minutes, from 12:08 to 13:08 CEST. Each line is the best of five date calls over SSH, each a new connection without ControlMaster; the offset is Pi time minus Mac time, the uncertainty half the round trip (ClockSync.measure).
| Minute | Offset (s) | Uncertainty (ms) |
|---|---|---|
| 0 | −14.947829 | 5.4 |
| 5 | −14.944454 | 6.0 |
| 10 | −14.942298 | 5.7 |
| 15 | −14.937529 | 8.0 |
| 20 | −14.937021 | 5.2 |
| 25 | −14.933902 | 5.4 |
| 30 | −14.938689 | 5.7 |
| 35 | −14.936051 | 6.2 |
| 40 | −14.934928 | 5.0 |
| 45 | −14.932568 | 5.4 |
| 50 | −14.930641 | 5.4 |
| 55 | −14.929101 | 6.0 |
| 60 | −14.927170 | 5.1 |
- The Pi was 14.95 s behind the Mac and ran 4.8 ppm fast: the least-squares slope over the thirteen points, 20.7 ms in the hour end to end, 0.42 s a day. The points lie within 3.9 ms of the line (rms 1.9 ms), inside the uncertainty of each.
- A session measures at its start and every 60 s and interpolates between the measurements (
ClockModel). At 4.8 ppm the clock moves 0.3 ms in 60 s, well within the ±5 ms of a measurement. A single measurement at the start would be 17 ms off after an hour. - The 14.95 s behind is what the RTC and the drift left since the clock was last set; this run does not show when that was.
- At 17:18 the same day, after the thirty warm reboots (a9000.md), a session measured the Pi 21.48 s behind, its eleven measurements within 9 ms of each other: 6.5 s further behind than in the hour, while the drift makes it gain. Each boot appears to lose part of a second, about 0.2 s on average, as when the clock is set from an RTC that keeps whole seconds. A session measures its own offset, so this does no harm.
The RTC without its time, 2026-11-02¶
A session at 22:33 measured the Pi 704,126.8 s behind the Mac, eight days and 3.6 hours, where a session at 01:37 that morning had measured 11.53 s. On the Pi at 22:53 by the Mac: up for 8 h 10 min, since about 14:43; its clock at 2026-10-25 18:18 UTC, System clock synchronized: no, and the RTC at 1970-01-01 08:10:27, which is the time since the power came back. The RTC battery had been unplugged once, and the Pi's power was cut before the boot (this boot's fsck found the boot partition's dirty bit set): without either, the RTC had lost its time and counted from zero, and systemd set the clock at boot to the time of /var/lib/systemd/timesync/clock, 2026-10-25 11:08:14, the last time systemd-timesyncd ran before linkside-system-setup masked it. The battery's charging is on (charging_voltage 3.0 V), and the battery read 3.01 V (vcgencmd pmic_read_adc BATT_V). The sessions measured their offset as before, within 13 ms of each other over 21 minutes, so their captures are corrected; the clock is set again from the Mac's (the sensor).