[LEAPSECS] real-time UT1 realisation

Warner Losh imp at bsdimp.com
Thu Sep 22 13:26:50 EDT 2011



On Sep 22, 2011, at 11:15 AM, Tom Van Baak wrote:


>> Eyeballing it says around a couple of milliseconds, which, unless

>> you have done things to your hardware, is lost in the NTP noise.

>

> That was my conclusion too. It's one thing to work with hardware

> at the micro- nano- or picosecond level. But the threshold with

> PC's, even PC's running NTP, is so relaxed that I figured almost

> any hack that uses a dut1 database would do the trick. The one

> I usually fetch is http://maia.usno.navy.mil/ser7/ser7.dat.

>

> Yesterday DUT1 was -0.30806, today is -0.30862, and tomorrow

> will be -0.30937. So it seems to me you could implement the

> google ntp server lie() function with essentially one line of code

> and it would be accurate to better than 1 ms. No interpolation,

> no filtering, no prediction; just a table lookup. Hard to beat that

> performance/simplicity ratio.


Yea, the hard part is keeping that table up to date, but I guess that's the job of a crontab entry not ntpd itself.

Warner



More information about the LEAPSECS mailing list