Watching Min, Max, Or Average Is Not Enough - Symbol WS5000 Series System Reference Manual

14-42
WS5000 Series Switch System Reference Guide
Figure 14.14 Distributions of received signal strength, as predicted by the average and standard deviationcalculated
across the collection of packets received.

14.6.1.1 Watching min, max, or average is not enough

Suppose your SLA (Service Level Agreement) states to provide a wireless signal strength of –63dB (or better)
to your customers partners/colleagues. What variables would you need to monitor to assure all involved that
the agreed service level was being achieved?
SLA 1 — End stations will experience –63dB or better
Monitoring min/max would not suffice. In most wireless infrastructures, this would tell you nothing as at least
one end station would have experienced –20dB (or some such excellent signal), and at least one end station
would have experienced –90dB (or some such signal at the very threshold of being detected).
Monitoring the mean would also not suffice.
end stations are experiencing a very similar level of signal ranging from –53dB to –63dB. In the blue case,
some are experiencing very strong signal while an equal number are experiencing very weak signal. Note that
in all cases the mean is identical.
Process control theory has addressed many such problems for decades using standard deviation. For any
normal distribution, 68% of the population resides in the range of mean +/– one standard deviation. 95% of
the population resides in the range of mean +/– two standard deviations, and 99.7% within three standard
deviations.
Note Six standard deviations include all but two billionths of the sample. Due to
the quality control methodology called 'Six Sigma', the term has acquired a
commonplace meaning as containing 99.99966% of the sample. Many
corporations have adopted the Six Sigma methodology for Quality Control and try
to achieve a defect rate of 3.4 per million. The rate of 3.4 per million, however, actually
corresponds to 4.5 standard deviations, because the Six Sigma founders assumed a natural
offset of 1.5 sigma to account for drift in production quality over time.
Figure 14.15 shows three possible scenarios. In the red case, all