Databases Reference
In-Depth Information
ered. With available applications ranging from complex data mining processes
to event detection, distributed systems have been applied in day-to-day op-
erations ranging from high performance computing networks to lightweight
and resource-constrained WSNs. To provide expandable coverage for different
types of applications, a dynamic and robust distributed pattern recognition
scheme must be able to perform under different network granularities.
2.6.1 Resource-Aware Approach
Resource considerations for conventional and distributed pattern recogni-
tion are distinctly different. In a distributed approach, the system must be
able to utilize the available resources effectively and e
ciently. To ensure
proper utilization and communication of resources between processing nodes,
a communication model needs to be considered.
Distributed pattern recognition (DPR) has the capability to scale up the
process when the size of the problem increases. However, scalability depends
on resource availability within a particular computational network. Resource
availability is influenced by the capacity and stability of the computational
network.
The network capacity of distributed applications, such as DPR, can be
viewed in terms of the granularity of the network. Commonly, computational
networks take the form of a coarse-grained network, such as grid computing, or
a fine-grained network, such as a wireless sensor network (WSN). The process-
ing capacity and capability of these networks may differ. Because application
deployment tends to focus on a single problem domain, most existing DPR
schemes are unable to adapt to different network granularities. Nevertheless,
some DPR schemes, such as the DHGN, have been developed with adaptive
network granularity considerations [4] and can be deployed in both coarse-
and fine-grained networks.
In addition to capacity, the stability of a computational network plays an
important role in determining its resource availability. A stable network is de-
fined as a network with minimal or no resource interference resulting from fault
or error occurrences. For a particular application, such as DPR, to perform its
function with minimal or no interruptions, a fault tolerance mechanism needs
to be considered.
2.6.2 Message-Passing Model in DPR
Process communication plays an important role in any distributed system
and determines how e
ciently a system adapts to different network configu-
rations and characteristics. Processes can be communicated within a network
using message exchanges between processing nodes. In any distributed sys-
tem, each processing node might require data exchange with other nodes to
complete a specific task or process. A thorough analysis of inter-process com-
munication must be performed to ensure that the proposed system is capable
