Database Reference
In-Depth Information
following form for each sensor:
v
ij
=
v
(
i−
1)
j
+
v
(
i−L
)
j
−
v
(
i−L−
1)
j
+
θe
i−
1
−
Θ
e
i−L
+
θ
Θ
e
i−L−
1
,
(2.3)
where
θ
and Θ are parameters of the SARIMA model,
e
i
are the predic-
tion errors and
L
is known as the seasonal period. For example, while
monitoring temperature,
L
could be set to one day, indicating that the
current temperature (
v
ij
) is related to the temperature yesterday at the
same time (
v
(
i−L
)
j
) and a previous time instant (
v
(
i−L−
1)
j
). In short,
the seasonal period
L
allows us to model the periodicity that is inherent
in certain types of data.
In the PRESTO system the proxies estimate the parameters of the
model given in Eq. (2.3), and then transmit these parameters to in-
dividual PRESTO sensors. The PRESTO sensors use these models to
predict the sensor value
v
ij
, and only transmit the raw sensor value
v
ij
to the proxies when the absolute difference between the predicted sensor
value and the raw sensor value is greater than a user-defined threshold
δ
. This task can be summarized as follows:
|
v
ij
−
v
ij
|
>δ,
transmit
v
ij
to proxy.
(2.4)
The PRESTO proxy also provides a confidence interval for each pre-
dicted value it computes using the SARIMA model. Like BBQ (refer
Section 2.3.2), this confidence interval can also be used for query pro-
cessing, since it represents an error bound on the predicted sensor value.
Similar to BBQ, the PRESTO proxy queries the PRESTO sensors only
when the desired confidence interval, specified by the query, could not
be satisfied with the values stored at the PRESTO proxy. In most cases,
the values stored at the proxy can be used for query processing, with-
out acquiring any further values from the PRESTO sensors. The only
difference between PRESTO and BBQ is that, PRESTO uses a differ-
ent measure of confidence as compared to BBQ. Further details of this
confidence interval can be found in [41].
2.4.2 Ken.
For reducing the communication cost, the Ken [12]
framework employs a similar strategy as PRESTO. Although there is a
key difference between Ken and PRESTO. PRESTO uses a SARIMA
model; this model only takes into account temporal correlations. On
the other hand, Ken uses a dynamic probabilistic model that takes into
account spatial and temporal correlations in the data. Since a large
quantity of sensor data is correlated spatially, and not only temporally,
Ken derives advantage from such spatio-temporal correlation.
The Ken framework has two types of entities,
sink
and
source
.Their
functionalities and capabilities are similar to the PRESTO proxy and the
