I run ntpd with a simple DCF77 receiver hooked up to a serial port as a reference clock (server 127.127.8.0 mode 14). This is precise to a few milliseconds. Overall accuracy depends on figuring out the delay introduced by the local receiver configuration. The PARSE/RAWDCF refclock driver defaults to 258 ms. When I set up this machine earlier this year, I determined the delay to be about 240 ms and compensated accordingly. (fudge 127.127.8.0 time1 0.240) A few days ago, I noticed accidentally that time is now off by about 0.8 seconds, compared to independent sources. $ ntpq -p remote refid st t when poll reach delay offset jitter ============================================================================== *GENERIC(0) .DCFa. 0 l 22 64 377 0.000 1.796 0.553 ntp1.arcor-onli 129.143.2.23 2 u 386 1024 377 14.816 804.110 2.812 151.189.13.46 129.143.2.23 2 u 444 1024 377 11.760 804.024 2.912 It looks like something has added a lot of latency to characters coming in the serial port. I have no idea when this happened. Candidates include the switch from sio(4) to uart(4), the new MPSAFE TTY layer, or maybe these are red herrings and it came with the ntpd update. Can anybody corroborate this? Any ideas? Hmm... 800 ms is a suspicious figure. Rx FIFO? -- Christian "naddy" Weisgerber naddy_at_mips.inka.deReceived on Sat Oct 04 2008 - 14:13:45 UTC
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:39:36 UTC