Database Reference
In-Depth Information
multi-path topology (and hence equipped to handle duplication). An
implication from the constraints is that the region running multi-path-
topology will finally be a sub-graph of the connectivity graph including
the base station. As shown in
Figure 3.1(c)
, the outer regions form trees,
while the region around the base station is multi-path-based. The nodes
at the boundary of the multi-path region are the switchable M nodes,
meaning that they can be switched to T nodes without violating the
correctness constraints. Also, the T nodes at the boundary of multi-path
region are called switchable T nodes, meaning that they can be switched
to M nodes freely. In order to adaptively adjust the topology of the
network according to the packet loss rates in different areas, an aggregate
on the data loss rate is maintained by each node
n
i
. Specifically,
n
i
computes the packet loss rate in its sub-tree. Once the packet loss rate
exceeds a user specifies threshold
, the sub-tree rooted at
n
i
suffers high
packet loss rate and applies the multi-path-based topology. Otherwise,
tree-based topology is applied. Changing between the two topologies is
accomplished by switching certain sensors between T and M nodes, so
that the multi-path region expands towards the areas with high packet
loss rate, while tree-based regions expand towards areas with low packet
loss rate.
4. Data Storage
Some applications do not involve a base station. For instance, often
scientists deploy sensors in the wild to monitor the habitat of animals
[2][3]. In such applications, the nodes form a WSN, which do not have
a base station. In order to collect data, scientists drive a vehicle with a
data collecting device through the monitoring territory. During the life
time of the network, the nodes store readings until they are contacted
by the collector. There are two main challenges for such applications:
i) due to the limited storage capacity, sensors memories may overflow
and ii) workload varies on different areas of the network, e.g., sensors in
the areas with frequent activities generate more data than those in areas
with rare activities. These issues raise challenges on how to store data
evenly in each node, and how to retrieve relevant data in different parts
of the network with low cost. [18] divides WSN storage techniques into
two categories: centralized and decentralized. In the centralized storage,
data are stored on the node that generates them. As an example, in
TinyDB, in order to perform certain kinds of aggregated queries, sensors
may store a small set of data locally [7]. This technique is not suitable
for an environment with frequent burst activities since they quickly drain
the valuable memory resource. A popular decentralized storage approach
