[ntp:questions] Using a uC based embedded device as SNTP server

Dave Hart hart at ntp.org
Fri Feb 17 16:29:27 UTC 2012

On Fri, Feb 17, 2012 at 01:16, Richard B. Gilbert
<rgilbert88 at comcast.net> wrote:
> If your Atmel uC gets its time from an atomic clock directly, it is Stratum
> 1.  If Atmel uC gets time from a stratum 1 server it is a stratum 2 clock
> and so on.  Each hop adds a little more garbage to the mix.

Stratum is far overrated as measure of accuracy.  In its early days,
NTP didn't have as careful accounting of hop-by-hop errors and it was
reasonable to assume higher stratum equals lower quality.  With
correctly-implemented NTPv4, this is much less so.  In fact, ntpd
4.2.7 has dramatically reduced the influence of stratum in clock
selection, addressing the problem of a high-delay distant lower
stratum source being relied upon to the exclusion of nearby higher
stratum sources with lower error statistics like root dispersion.  It
appears to work well, even when the sources are using NTPv3 or early
ntpd 4.x releases with less accurate error budgeting.

You are of course correct in saying the embedded device is only
stratum 1 with a local reference clock, though it need not be a
particularly high-quality one.  time-nw.nist.gov and a number of other
NIST stratum 1s use the ACTS telephone time service, for example,
which has been compromised by the increasing transition of
long-distance telephony away from synchronous, fixed delay transport
to packet-based carriage.

Dave Hart

More information about the questions mailing list