\reviewtitle{A Taxonomy of Sensor Network Communication Models}
\reviewlabel{tilak02taxonomy}
\reviewauthor{S. Tilak, N. B. Abu-Ghazaleh and W. Heinzelman}

The authors provide a good overview of sensor utilization and, in
particular, of how network protocols are used to communicate between
sensors and the observers.  They comment that sensor applications have
different goals from traditional wireless networks.  In a sensor
network, one or more observers use sensors, which consist of sensing
hardware, memory, a battery, an embedded processor, and a
trans-receiver, to receive information about one or more phenomena.

System lifetime ``can be measured by generic parameters such as the
time until half of the nodes die or by application-directed metrics,
such as when the network stops providing the application with the
desired information about the phenomena.''  ``To gain more accuracy, the
application can either request more frequent data dissemination from
the same sensor nodes or it can direct data dissemination from more
sensor nodes with the same frequency.''  Data replication can be
achieved by, \eg the SPIN protocol.

The network protocol used ``is responsible for supporting all
communication, both among sensor nodes as well as between the sensor
nodes and the observer(s).''  This pair of responsibilities allows
them to classify communication into two categories: application and
infrastructure.  They describe cooperative and non-cooperative
application communication, but I am unclear what non-cooperative is
and why is would ever occur.  Satisfying application communication may
require reconfiguring the network.  Infrastructure can be thought of
as overhead (and should be minimized).  Clustering and selecting a
cluster-head is infrastructure communication.

There are three types of data delivery, which depend on what is
required by the application: continuous, event-driven, and
observer-initiated.  There are hybrids of these types.

How packets of information are actual sent occurs through flooding
(broadcast-based), unicast, and multicast.  In multicast ``sensors
form application directed groups and use multicast to communicate
among group members.  The observer could communicate with any member
of the group to obtain the desired data.''

Setting up paths of communication is somewhat similar to the problem
of routing in ad hoc networks.

In static sensor networks ``localized algorithms'' that often use a
hierarchy have been shown effective.  In dynamic networks, either the
sensors, the observer, or the phenomenon are mobile.  To repair paths
of communication, there are both reactive approaches that try to
repair breaks only after they occur and proactive approaches that
anticipate future broken paths.  One approach to mobile phenomena is
the Frisbee model.  DSDV is a proactive approach for use under
continuous data delivery.  LEACH uses a local, rotating cluster-head
approach.  ``Directed Diffusion (DD) is a data-centric protocol, where
nodes are not addressed by their addresses but by the data they
sense.''  Garcia-Molina has proposed a publish/subscribe approach.