Information Technology Reference
In-Depth Information
provides a distributed fault tolerant approach for
regional event extraction in sensor networks. All
nodes with readings of interest in a given vicinity,
i.e., the region of event, are formed into a cluster
where the node with the lowest id-number becomes
the cluster head. The cluster head collects all read-
ings and performs a majority decision. Since their
approach is based on binary event detection where
all nodes signal a “Yes” or “No” instead of send-
ing their measurements, the cluster head simply
counts all statements. If more than 50 percent of
the participating nodes state a positive event, the
cluster head forwards this event to the central
sink in the network. Beside performance issues,
they do not take care on energy resources because
large event regions produce much overhead for
communication.
Krasniewski et al. proposed TIBFIT (Kras-
niewski, Varadharajan, Rabeler, Bagchi, & Hu,
2005), a protocol able to cope with arbitrary data
faults and malicious nodes. It shall enable reliable
and fault tolerant data gathering by assigning trust
values to the sensor nodes. These values confirm
the plausibility of correct measurements or state
a lack of credibility for single nodes. The head
node of a cluster collects the readings and trust
values of all nodes and decides whether an event
has occurred or not. Due to the decision of the
head node, the trust values of all correct reporting
nodes increase whereas the other values decrease
respectively. To make sure the trust values are
correct, at least two shadow head nodes monitor
all activities and results of the decision process
in background and take corrective action if nec-
essary. TIBFIT achieves a good fault tolerant
performance even if more than 50 percent of the
sensor measurements are faulty, provided that the
initial monitoring phase is long enough to establish
the trust values. Additionally, this protocol is able
to cope with malicious nodes, which can only
temporary influence the decision process because
their confidence values decrease with every faulty
report. This is only true if the number of malicious
nodes is less than the number of correct reporting
nodes. Unfortunately, required overhead was not
measured or calculated. However, the algorithms
used for collecting and distributing sensor readings
and trust values allow assumption of an enormous
overhead, especially for the usage of shadow head
nodes. Hence, the efficiency of the provided fault
tolerant performance strongly depends on the ap-
plication it is used for.
Kamiya et al. (Kamiya, Mineno, Ishikawa,
Osano, & Mizuno, 2008) applied a P2P network
of sensor gateways to maintain event detection
across several heterogeneous sensor networks.
Each sensor gateway accesses and manages a
certain sensor network. To define event detec-
tion in one or more maintained sensor networks,
the sensor gateways provide an XML event de-
scription parser that splits complex events into
required atomic ones and registers these at the
corresponding sensor gateways. The underlying
sensor networks continuously report their raw
sensor readings to the gateway nodes, which fi-
nally evaluate the atomic and respective complex
events. Even here the sensor gateways constitute
a SPoF. Just like all other discussed approaches
relying on gateway or centralized nodes, this
is again very inefficient with regard to energy
consumption and network load. Due to the fact
that atomic events are not forwarded to the actu-
ally measuring sensor nodes for evaluation, all
sensor readings need to be sent to the gateway
nodes even in case of no event is triggered. By
that the energy consumption is far from optimal.
Nevertheless, autonomous management of sensor
networks based on an event description parser is
a promising approach to reduce the complexity of
node and network configuration. Unfortunately,
the XML descriptions used at the parser are not
presented making it impossible to draw conclu-
sions about their applicability.
The Context Dependent Event Detection
(CoDED) platform (Schwiderski-Grosche, 2008)
presents an architecture for context-dependent
event detection in sensor networks. In order to
save energy resources, events are monitored in
Search WWH ::
Custom Search