[ntp:questions] Re: Philosophical question about strata

David L. Mills mills at udel.edu
Fri Oct 14 17:29:22 UTC 2005


Really and truly do believe interpreting stratum as quality of service 
is a bright red herring. It was never intended for that purpose. Its 
primary purpose is avoiding timing loops. There is an absolutely 
wonderful metric with which to interpret quality, a combination of 
maximum error (synchronization distance) and estimated error (system 
jitter). There are explicit provisions in the reference clock interface 
that allow the driver to adjust these values with respect to whatever 
statistics the clock provides. It would seem the PDoP commonly provided 
by the GPS receiver firmware would be a prime candidate.


Richard B. Gilbert wrote:

> Wolfgang S. Rupprecht wrote:
>> "Lee Sailer" <lee.sailer at direcway.com> writes:
>>> 1.  If everything is working, stratum 1 is better than stratum 2.
>>> 2.  If the stratum 1 hardware is suffering a slow degradation
>>> breakdown, then the stratum 2 will be better.
>> Well, this is a sad (but true) commentary on the state of consumer gps
>> firmware.  The gps (x,y,z,t) calculations, if done correctly, will
>> tell you when things are going badly.  One only needs to see 4
>> satellites to work out the 4 unknowns.  Usually one sees 8-12.  There
>> is an awful lot of redundant information in observing those extra
>> satellites that one can use to determine error bars for the calculated
>> x,y,z and time.  There really isn't a good reason why a gps can't tell
>> you "Hey, these numbers are out to lunch, please disregard."
>> If the failing gps's in question put out EPE, HDOP or similar "quality
>> of observation" numbers, it might be good to double-check that the ntp
>> drivers really do ignore once the readings get bad.
>> Question to the hard-core ntp jocks: Would anything break within the
>> internal of ntp if the stratum of a reference clock were to be
>> forcefully changed from the traditional stratum-1 for a correctly
>> running reference to, say, stratum-5 for a gps with really crummy EPE
>> that might be giving good time, but then again might not?  (I assume
>> this is within the spirit of "stratum" with stratum being an
>> estimation of goodness, but please speak up if this isn't so.)  I'm
>> guessing that changing the stratum of an ntp reference clock should
>> just work, because remote servers/peers can change their stratum and
>> ntp automatically adapts to the new landscape.
>> -wolfgang
> I don't think of stratum as an "estimate" of goodness.   I think it's 
> purely a designation of position in the hierarchy.  A stratum one server 
> is stratum one because it gets its time from a primary standard; e.g. an 
> atomic clock.   A server that gets its time from a WWV receiver is 
> technically stratum one and can be several milliseconds off because of 
> the vagaries of HF radio propagation.   The "goodness" of a server also 
> depends on the path through which you receive time from it.   A client 
> that is three thousand miles away from a stratum one server and 
> receiving time over a heavily used network  is probably getting time 
> that is an order of magnitude poorer than a client three hundred feet away.
> Fudging a server to a higher stratum than it would normally have should 
> make it appear less desirable to any client that has a choice of servers.

More information about the questions mailing list