High Level Definition of Event-Based Applications for Pervasive Systems - Advanced Design Approaches to Emerging Software Systems - page 167

Figure 14. Comparison of required messages in the entire network in application of lease-based and
ACK-based collaboration in case of permanently failing sensing capabilities. Despite the significantly
enhanced detection performance, the lease-based approach only requires to transmit 0.35 messages per
node and interval. Even using the reliable mode in lease-based collaboration, which does not influence
the detection results, requires to transmit only about one message in two intervals. In contrast to that,
the ACK-based approach required in average more than 7 messages per node and interval. This caused
the simulation runs to be aborted by the simulator. Please note, the diagram applied a logarithmic scale.

introduced deployments considering transiently
failing devices and completely failing nodes, too.
From these simulations we learned fundamental
aspects about collaborative event detection. Col-
laboration can significantly enhance the robustness
of a sensor network. It keeps on running its applica-
tions with high detection accuracy even in case of
failed sensing devices. The presented ACK-based
collaboration scheme provides the best detection
accuracy since it refreshes the actual values of
EDT-nodes after each interval. But simultaneously
it requires a huge number of collaboration mes-
sages. It was shown that the cost-efficiency of the
lease-based approach is very high and reduces the
number of collaboration messages by a factor of
10 or higher. The lease-based publish/subscribe
collaboration scheme can be configured to such
extent that it is able to achieve detection accura-
cies that closely meet those of the ACK-based
scheme by choosing a proper leasing time. This
leasing time primarily depends on the expected
properties and behavior of the phenomenon to be

sensed. The leasing time ideally is less or equal
to the mean time of exposure to the monitored
phenomenon. In the simulations, this was a leas-
ing time of one minute (six intervals). In addition,
changes that influence the event detection have to
be considered, too. Such changes are caused by
failed sensing devices and crashed or moved sen-
sor nodes. By that, publisher nodes may disappear
or become unable to publish further EDT-node
values. Therefore, the estimated mean time to
failure has to be regarded, too. For configuration
of a proper leasing time, the user has to obey both
of the following restrictions:

1. The maximum leasing time is less or equal
to the mean time of exposure to the phenom-
enon to be sensed.
2. The maximum leasing time is less than the
mean time to failure.

Of course, the lower bound of a leasing time
is one EDT evaluation interval. In that case, the