Re: [gpsd-dev] [PATCH] Improve chrony examples in time howto.

[Greg Troxel]

Miroslav Lichvar <address@hidden> writes:

> On Mon, Nov 04, 2013 at 10:03:29AM -0500, Greg Troxel wrote:
>> "Eric S. Raymond" <address@hidden> writes:
>> > Miroslav Lichvar <address@hidden>:
>> >> On Mon, Nov 04, 2013 at 07:18:16AM -0500, Greg Troxel wrote:
>> >> > Miroslav Lichvar <address@hidden> writes:
>> >> Please note that the performance of chronyd on non-Linux systems may
>> >> not be adequate for the precision of the PPS source. Its system driver
>> >> will use only the old adjtime() system call to discipline the clock,
>> >> there is no correction of the frequency offset and the resolution is
>> >> limited to microsecond.
>> >
>> > Ship another patch to this effect, please.
>> 
>> And please be more measured about "adequate".  The notion that there
>> are sub-microsecond issues can be a bit hysterical on systems that don't
>> have all other issues addressed.
>
> Any suggestions on how to rephrase it? The problem is the uncorrected
> frequency. The time error of the clock has a sawtooth profile with
> amplitude proportional to the drift of the clock. If the drift is in
> single digit ppm, perhaps the error may be acceptable, but I'm not
> really sure.

I would try to explain it clearly and be quantitative.  There is a trend
in timekeeping to be horrified at the last ns, even though hardly anyone
comes close; I think people should be aware of errors and manage them as
an engineering exercise).

IF you mean chrony uses RFC1589 ntp_adjtime on Linux, and not on other
systems, that sounds buggy, as other systems have ntp_adjtime.

pgpssFluI_hRe.pgp
Description: PGP signature