Image Processing Reference
In-Depth Information
a subset of these paths. Reliable paths are usually reinforced while unreliable ones are removed by
expiration due to lack of reinforcements or explicit negative reinforcements. Such gradients allow the
local repair of failed or degraded paths and do not require the reflooding of the interest. However, it
is necessary to perform flooding when a new interest is injected into the network.
While directed diffusion performs routing based on named data, TinyDB performs routing using
a semantic routing tree (SRT) which is based on the actual values of sensor readings. It is useful for
servicing range queries. An SRT is an index built over some constant attribute
A
andisstoredlocally
at every node in the network. he index at a node consists of a one-dimensional interval representing
the range of
A
values being generated not just by the node itself but also by all its descendants. When
a node encounters a query, it only forwards it to its immediate children which are reported to be
transmitting values matching the required range specified in the query. he readings may have been
generated either by any of the immediate children or by any of the nodes within the subtrees rooted at
the immediate children. Additionally, a node executes a query by itself if it can be serviced locally and
subsequently transmits the result to its parent. The result eventually propagates up the tree toward
theroot.Ifthequerycannotbeservicedbythenodeoranyofitschildren,itisdropped.heentryof
a child node expires from the SRT of a parent node if the parent node does not receive any updates
from the child within a predefined timeout period. he parent then updates its interval information
by querying its children and also informs nodes higher up the hierarchy if any changes are detected.
While the SRT maintenance algorithm is capable of reflecting changes in the network dynamics (e.g.,
deathofanode),thecostofupdatingrangescouldbeprohibitiveifthevalueofthemeasuredattribute
changes too rapidly.
Query and data dissemination schemes that prevent the need to flood the entire network have
progressed markedly in the recent past. From initially just considering event types or ranges of actual
sensor readings, they currently support multiple range queries and use various hashing functions to
direct queries to the appropriate sections of the network using the attribute types and values as inputs.
Furthermore, they incorporate in-network storage as an integral part of the query dissemination
mechanism. However, these newer schemes assume that all nodes are location aware thus increasing
the complexity of the system.
4.4.7 Concluding Remarks
As time progresses, WSN deployments are gradually going to grow larger and certain deployments
may even be enlarged in stages. his makes it increasingly necessary to improve support for heteroge-
neous networks, multiple roots and optimization of multiple simultaneous queries that may overlap
partially over sensor types, readings, and spatial and temporal parameters. Cross-layer optimizations
from the application layer also need to dig in deeper into the network layers and attempt to eventu-
ally influence the operation of the MAC. Query optimizations need to be pushed into the network to
prevent large amounts of metadata being sent back to the root.
4.5 Conclusion
In this chapter we have outlined the characteristics of WSNs from an architectural point of view. As
sensor networks are designed for specific applications, there is no precise architecture to fit them all
but rather a common set of characteristics that can be taken as a starting point.
The combination of the data-centric features of sensor networks and the need to have a dynamic
reconfigurable structure has led to a new architecture that provides enhanced capabilities than
the existing ones. We highlighted the characteristics of the new architecture—for a more detailed
description, the user is referred to Refs. [,].
