High Level Definition of Event-Based Applications for Pervasive Systems - Advanced Design Approaches to Emerging Software Systems - page 162

same EDT node within a certain detection interval.
Hence, the subscriber has to determine the correct
value of that EDT node, or more precisely, the value
that is most likely correct. To cope with that, the
sensor node counts all true and false values per
interval and finally decides on the majority. The
respective counters at each EDT node are reset
to zero after every evaluation run. In case of a tie
situation, the last received value is taken.
Data security is a side condition that has not
been considered so far but is an important issue
for distributed application. This chapter does
not focus on ensuring data security. But as it is
essential with respect to reliability, especially
when considering mission- and safety-critical
applications, some means providing data security
must at least be mentioned. Using wireless com-
munication increases the risk of malicious attacks
because the sensor network becomes accessible
from outside. Whereas dozens of various attacks
exist, the integrity of collaboration messages is of
particular interest. The integrity and confidential-
ity of exchanged information can be assured by
secure hashing and encryption techniques. These
means can additionally be used on top of the
presented event detection scheme, if necessary.
Many different encryption methods are already
available, but these quite differ in their strength and
energy consumption depending on the algorithm
and hardware used. Just as mentioned before,
the associated overhead associated to security
mechanisms is to be balanced between varying
application requirements, too.

subject to failures in sensing devices. Based on that,
application of the lease-based publish/subscribe
approach is compared to the ACK-based variant.
The simulation results are evaluated with regard
to the detection accuracy and cost-efficiency of
the introduced algorithms.
Application of Boolean event detection allows
to generate compact snapshots of the system state
containing the actual or respectively last detection
results of all nodes, which are either event or no
event respectively 1 or 0 . As a start, the first run
simulated the correct behavior of the entire sensor
network without any failures while snapshots from
the system are frequently generated at every EDT
evaluation interval. These snapshots represent
the best case scenario and are used as the regular
reference. In the following, all results and evalu-
ations based on these snapshots are henceforth
called reference . Consequently, such snapshots
were taken in all other simulation runs at equal
simulation time, too. Finally, the snapshots of
equal simulation time are matched against the
reference to determine whether the sensor nodes in
the simulation runs gathered the correct detection
result or not. The simulation results are not only
compared to the reference scenario. Each failure
scenario was also executed without improvement
by collaboration to gather the usual local detection
results of the sensor nodes. In the following, these
runs are referred to as standard .
The simulation results are analyzed to deter-
mine the total detection accuracy and the number
of required messages. The total detection accuracy
states the number of correctly gathered positive and
negative evaluation results per interval. Hence, a
detection accuracy of 100% is given if all nodes
within the phenomenon notify an event (positive
result) while all other nodes do not register an event
(negative result). As mentioned, the collaboration
algorithms introduce a communication overhead
represented by the total number of sent overhead
messages. Additionally, the average number of
required messages per node and interval is de-

Performance Evaluation
Methodology

To prove the concept, a prototype of the EDT
was implemented in the discrete event simulator
OMNeT++ (www.omnetpp.org) (Varga, 2002)
with an extension for simulation of WSNs called
Castalia (http://castalia.npc.nicta.com.au/) (Pham,
Pediaditakis, & Boulis, 2007). The introduced fire
detection scenario was applied to WSNs that are