High Level Definition of Event-Based Applications for Pervasive Systems - Advanced Design Approaches to Emerging Software Systems - page 165

Table 3. Comparison of applying lease-based publish/subscribe and ACK-based collaboration in case
of permanently failing sensing capabilities. The lease-based approach performs best and enhances the
standard detection by about 30%. Due to the successively increasing number of necessary collaboration
messages in the ACK-based scheme, the simulation process has been killed by the simulation environ-
ment. For the affected runs the last known system state is represented and marked with an *.

Total detection accuracy in %

Figure 13 Standard Lease = 6 ACK

Uniform random deployment * 63.970 82.917 85.619

Uniform random deployment 59.765 77.968 -

Average number of collaboration messages per node and interval
Figure 14

Lease = 6 (reliable) Lease = 6 ACK

Uniform random deployment* 0.537 0.357 7.360

Uniform random deployment 0.525 0.356 -

The total detection accuracies of both collabo-
ration approaches are compared to the standard
detection in Figure 13. In case that 50% of the local
detection results are lost in the standard detection,
the lease-based and the ACK-based scheme still
provide a detection accuracy of 82% and 85%
respectively. If 70% of detection results are lost in
the standard detection, i.e., only 30 nodes remain
functional, the collaboration schemes still generate
67% (lease-based) and 70% (ACK-based) correct
detection results. That is a temporarily detection
improvement of 225% compared to the standard
detection. The lease-based approach significantly
outperforms the ACK-based variant in comparison
of the number of collaboration messages, see the
diagram in Figure 14. It requires only about one
message within three detection intervals. The
ACK-based scheme sent about 7.5 messages per
node in average at each interval, which in total
differs from the lease-based scheme by a factor
of 20. This is obviously too much traffic to be
simulated. It further limits the applicability of
the ACK-based scheme with respect to the node
density of the network and the amount of data
that needs to be exchanged.
Only the lease-based approach and the standard
detection have completed their simulation runs.
The following compares the respective perfor-

mances. Again, the lease-based approach en-
hances the total detection accuracy in the network
by an average of 30% in the uniform random
deployment. It further detected 82% of all exist-
ing phenomena, which represents a gain of 39%
in comparison to the standard method. This sig-
nificant increase is achieved by requiring a col-
laboration overhead of only 0.35 messages per
node and interval. This is equivalent to the trans-
mission of 35 messages in the entire network
during one detection interval representing ten
seconds in lifetime. Even using the reliable mode
for lease-based collaboration, i.e., to explicitly
confirm each published value, increased the
overhead to 0.53 messages per interval only. The
reliable publishing does not influence the detec-
tion accuracy. It merely triggers explicit acknowl-
edgement messages for received publications.
An increasing number of unavailable sensing
devices continuously decreases the performance
of event detection of course. In contrast to the
standard detection, both introduced collaboration
schemes significantly extend the time of running
the detection with high detection accuracy. Even
if 50% of all sensor nodes cannot evaluate the
EDTs with own sensor readings, both collaboration
schemes still provide a total detection accuracy
of at least 84%. As expected, the ACK-based