Am Sonntag, 10. Januar 2021, 17:30:52 CET schrieb Ben Hall: > While on the NTP-Sec server, on the output of ntpq, there are at least four: > > delay offset jitter > ====================== > 0.0000 0.0740 0.0027 > 85.4951 -1.7276 11.7616 > 40.9579 -1.9249 13.8848 > 25.7753 -0.5499 16.9417
All modern OS that NTPsec runs on provide nanosecond resolution. The prcision and accuracy of those measurements vary with the hardware, the rasPi in particular is limited to something on the order of 150ns granularity as it runs the GPIO off a 19.2MHz clock. > Is this just trickery or is that real precision? I don't know. Here's the statistics from the NTP loop data of my NTPsec boxes for the last seven days: tb1 PLL/FLL: bias = +1.284e-10s, mabs = 1.362e-07s (de-biased) tb2 PLL/FLL: bias = +5.827e-10s, mabs = 2.510e-07s (de-biased) rpi2 PLL/FLL: bias = -6.607e-11s, mabs = 2.507e-07s (de-biased) rpi3 PLL/FLL: bias = -2.640e-09s, mabs = 1.665e-07s (de-biased) rpi4 PLL/FLL: bias = -1.397e-09s, mabs = 3.081e-07s (de-biased) rpi5 PLL/FLL: bias = -1.058e-09s, mabs = 2.743e-07s (de-biased) These are all self-ovenized so they do run close to the measurement resolution. One day I'll get around to try polling instead of interrupts to get rid of some of the noise and/or up the frequency on the GPIO, the hardware would allow 250MHz there… These three are non-ovenized and have at least 10x worse performance: rpi1 PLL/FLL: bias = -1.246e-07s, mabs = 1.552e-05s (de-biased) gertrud PLL/FLL: bias = -6.350e-07s, mabs = 6.341e-06s (de-biased) otto PLL/FLL: bias = -1.541e-07s, mabs = 1.456e-05s (de-biased) So, for a "normal" setup I'd trust maybe 10 microsecond figures, but you can get it down into the 100 nanosecond range with some care. > While the Win10 box has yet to reach the Teensy 4.1-based NTP server, > the RPi has no issues, likes what it sees, and uses it for time-sync > without problem: > > delay offset jitter > ======================= > 0.076 1.180 0.261 The jitter seems pretty high and the delay is oddly low. Has the NTP loop actually converged already? If you use a rasPi with Gigabit Ethernet make sure you switch off any link power management and set a fixed CPU frequency. Regards, Achim. -- +<[Q+ Matrix-12 WAVE#46+305 Neuron microQkb Andromeda XTk Blofeld]>+ Factory and User Sound Singles for Waldorf Blofeld: http://Synth.Stromeko.net/Downloads.html#WaldorfSounds _______________________________________________ time-nuts mailing list -- [email protected] To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com and follow the instructions there.
