Typically there are three
main categories of performance data presented
to you by Performance Management products:
measured, calculated and modeled. Be very
wary of any products that try to blur the
lines between the three!
Measured data offers values that have been
measured directly from the operating system.
Third party performance products obtain
this data from either the operating system
meters or from utilities supported by the
manufacturer that themselves take the data
from the meters. As both sources are the
same, each is as accurate an approach as
the other. The latter benefits from quicker
and easier porting between operating system
releases for the supplier. Also, the utilities
approach means that the third party is not
introducing additional risk by loading software
onto your machine that accesses the operating
system kernel.
Calculated data are values derived from
measured data. This is done to 'fill in'
gaps where the operating system does not
produce the metrics required. Clearly the
relevant screens or associated Help text
should state what values are measured and
what are calculated. Often calculations
rely on assumptions made by the third party
supplier. The quality of these assumptions
directly affects the quality of the data.
Any supplier that won't 'come clean' about
where and what its assumptions are must
naturally be treated with some suspicion!
Even worse, if they can't come clean!
modeled data relates specifically to capacity
planning products that use the 'analytical
modeling' approach to performance prediction.
This technique is invaluable in predicting
relative change in response time for a given
change in the environment e.g. hardware
upgrade or workload increase. Put alongside
measured data without explanation however,
and it is thoroughly misleading. As managers
often want response time data, you could
end up embarrassed if you mistakenly give
them figures purporting to be measured -
especially if the predicted values don't
match user experience!
The rules are simple. Know where the data
you are using comes from, and don't trust
a supplier that tries to hide the facts!
Next
modeling Tip |