Database Reference
In-Depth Information
of their techniques, they can only provide probabilistic guarantees on
the accuracy of the data that the sink collects, and hence no absolute
bound on the error. While this may be su
cient for certain applications
(e.g., temperature and humidity monitoring for Heating, Ventilation and
Air Conditioning systems), there exists a class of applications, for which
hard accuracy guarantees are essential (e.g., scientific applications that
need accurate, fine-grained monitoring of some phenomenon).
In several scientific applications, it may also be the case that the
domain experts do not already have a model of the data distribution
they are sampling using the WSN, but are rather interested in collecting
accurate measurements in order to build such a model [19]. Indeed,
WSNs offer a unique opportunity to scientists to observe phenomena
and develop models for them at a scale and granularity that were never
before possible. Nevertheless, in order to so, they need to have accuracy
guarantees on the sensor measurements.
In data-driven data acquisition, we make the assumption that the ap-
plication running at the sink allows for a small tolerance in the accuracy
of the reported data. In contrast with the ideal requirements of the sink
obtaining
exact
values in
all
data reports, the correctness of these ap-
plications is unaffected as long as
i)
the reported values match
closely
the exact ones;
ii)
inaccurate values occur only
occasionally
.Inother
words, deviations from the exact reports are acceptable, as long as their
extent in terms of difference in
value
and
time interval
during which
the deviation occurs are small enough. We capture these assumptions,
common to many applications, with the following definitions on value
term
error tolerance
,
ε
VT
to refer to both of them together.
Definition 7.1 (Value Tolerance)
Let
V
i
be an exact measurement
taken at time
t
i
.The
value tolerance
is defined by the maximum relative
and absolute errors acceptable,
ε
V
=(
rel
,
abs
)
. From the application
perspective, reading a value
V
i
becomes equivalent to reading any value
V
i
in the range
R
V
defined by the maximum error,
V
i
∈
R
V
=[
V
i
−
,V
i
+
]
,
V
i
100
rel
,
abs
where
=max
{
}
. In other words, the application considers
avalue
V
i
∈
R
V
as
correct
.
Note that the value tolerance includes both an absolute and a relative
component. This is useful for applications that involve sensor readings
with wide ranges of values.
Definition 7.2 (Time Tolerance)
Let
T
=
|
t
j
−
t
k
|
be a time inter-
val, and
V
T
=
V
j
,..., V
k
}
the set of values reported to the application
during
T
.The
time tolerance
ε
T
is the maximum acceptable value of
{
