Guys,
Is there a misunderstanding here? The time available to all system
components ultimately comes from an atomic variable maintained by the
kernel, usually in seconds and nanoseconds. The time and frequency of
this variable is disciplined (sic) by NTP, either via another server or
by a local reference clock. With respect to NTP, the precision is
defined as the time to read the clock, usually the gettimeofday() or
similar system call. In modern systems the precision is typically less
than one microsecond. Note that the precision of the NTP timestamp
itself is 232 picoseconds, or about 0.2 nanosecond.
You ask about the accuracy of an NTP time server. This depends on a
number of factors, machine room temperature variations, network delay
variations, etc. With a GPS receiver and pulse-per-second (PPS) signal,
typical accuracies are in the low tens of microseconds. Clients sharing
a common Ethernet with such a server typically show errors less than 100
microseconds. Conventional wisdom suggests better than 1 ms for campus
networks, better than 10 ms with fast Internet connections and better
than 50 ms in all but really broken paths.
Much more information is available at www.ntp.org.
Dave
Post by J de Boyne PollardFC> Furthermore, even synchronized time is not synchronized
FC> to the precision of NFS timestamps. So not only is it not
FC> realistic to assume "precisly synchronized time" (relative
FC> to timestamp granularity), it is impossible.
SR> How accurate are network time servers?
Are you asking how _accurate_ they are or how _precise_ they are?