NetBSD/dist/ntp/html/measure.htm

51 lines
2.6 KiB
HTML

<html><head><title>
Time and Time Interval Measurement with Application to Computer and
Network Performance Evaluation
</title></head><body><h3>
Time and Time Interval Measurement with Application to Computer and
Network Performance Evaluation
</h3><hr>
<p>The technical memorandum: <cite>Time and Time Interval Measurement
with Application to Computer and Network Performance Evaluation</cite><a
href="http://www.eecis.udel.edu/~mills/database/memos/memo96a.ps">
(PostScript) </a> describes a number of techniques for conducting
experiments typical of computer network and transmission systems
engineering.
<p>In most experiments in which time is involved, it is necessary to
develop estimates of time, frequency and measurement errors from a
series of time measurements between the clocks of a number of computers
and ancillary devices interconnected by some kind of computer network.
However, time is not a physical quantity, such as mass, nor can it be
measured relative to an absolute frame of reference, such as velocity.
The only way to measure time in our universe is to compare the reading
of one clock, which runs according to its own timescale, with another
clock, which runs according to a given timescale, at some given instant
or epoch. The errors arise from the precision of time comparisons and
the accuracy of frequency estimates between the timescales involved.
<p>The usual data collected during a performance run of some experiment
might include time offsets, time delays, frequency offsets and various
error statistics. While time offsets between two clocks can be measured
directly, frequency offsets can be estimated only from two or more time
offsets made over some time interval in the experiment. In practice, a
sequence of time comparisons can be performed over the lifetime of the
experiment and the instantaneous frequency estimated either in real time
with a recurrence relation, or retrospectively with a polynomial fit to
the data.
<p>Estimating time and frequency errors in real time has been studied by
a distinct subspecies of physicists who have made a career of the
technology involved. Various means including autoregressive models,
Kalman filters and simple weighted-average algorithms are used
extensively by national standards laboratories to model cesium-clock
ensembles. These techniques have been adapted to computer network and
transmission engineering problems as well. This memorandum explores
issues in performing experiments of this type and summarizes various
techniques found useful in practice.
<hr><a href=index.htm>Home</a><address><a
href=mailto:mills@udel.edu> David L. Mills &lt;mills@udel.edu&gt;</a>
</address></a></body></html>