High Level Definition of Event-Based Applications for Pervasive Systems - Advanced Design Approaches to Emerging Software Systems - page 163

termined to directly compare the communication
efficiency of all approaches.

radius of 1.5 meters. In this area all sensor readings
clearly exceed the defined thresholds. The area
with radius between 1.5 and 3 meters specifies the
immediate vicinity of the fire centre, where the
sensor readings slightly exceed the thresholds and
the phenomenon can still be recognized. The area
with radius from 3 to 6 meters defines the outer
expansion of the phenomenon. In this area the
sensor readings are slightly below the thresholds.
Finally, all nodes not located in one of these three
areas of the phenomenon generate “usual” sensor
readings that are clearly below the thresholds.
To trigger changes in sensor readings and node
evaluation results, the centre of the fire phenom-
enon deterministically moves within the network
boundaries at every minute or six EDT evaluation
intervals respectively. In comparison to the spatial
expansion covered by the entire sensor network the
size of the phenomenon is rather small. Most sensor
nodes will therefore generate negative evaluation
results per interval whereas only a few nodes may
possibly detect the phenomenon. This perfectly
fits to the fire detection scenario where upcoming
fires usually feature a small size. However, such
moving behavior does not correspond to a fire in
the real world. Nevertheless, the introduced fire
detection scenario is a well descriptive vehicle
to exemplify phenomenon definition and reliable
phenomenon detection within a mission-critical
context. The described phenomenon is used to
generate deterministic sensor readings and simu-
lation results. The case that sensor nodes may be
damaged or destroyed by such phenomenon is
also not considered in this scenario.

Simulation Parameters and
Deployment Patterns

According to the introduced event specification
of the fire event, an EDT evaluation interval is
ten seconds. The simulations ran three simulated
hours, which is equivalent to 1080 EDT evaluation
intervals in the fire detection scenario. In the fol-
lowing, the simulation time is given in discrete time
steps, i.e., the EDT evaluation intervals. Finally,
the simulation parameters regarding the wireless
communication need to be identified. Wireless
communication is subject to many restrictions
resulting in an unreliable and sometimes nonde-
terministic performance. Many research projects
already studied the parameters of link reliability,
end to end delays, low power communication
etc. These issues are indeed important, but were
not considered in our simulations. These applied
ideal conditions at the MAC layer and the wire-
less channel to generate deterministic results for
comparison. Simulation results are taken from
ten different random uniform node deployments
using a field of 22.5×22.5 meters containing 100
wireless sensor nodes. The average results of all
simulation runs on each of the ten deployments are
determined. Each sensor node initially possesses
all required sensing facilities, i.e., carbon mon-
oxide, temperature and smoke detectors. Hence,
all sensors are initially enabled to locally evaluate
the complete EDT to gain local detection results.
For deterministic event generation, a simu-
lated phenomenon is specified, which causes
the actual sensor reading in a certain region. The
simulated fire phenomenon had a circular dimen-
sion specifying high sensor readings in its centre,
which decrease with the distance to the centre of
the phenomenon. In particular, this phenomenon
partitions the network into four areas. The region
of the fire producing the highest sensor readings is
defined as a circle around the centre point with a

Failure Scenario: Permanently
Failing Sensing Capabilities

Low cost production, decreasing energy supply
and various environmental influences may not only
lead to errors of measurement in sensor readings.
These also cause sensing devices to fail transiently
or to get even permanently lost. In that case, usual
local event detection based on own sensor read-