Databases Reference
In-Depth Information
FIGURE 8.3: Proposed commodity grid-based distributed pattern recognition
framework.
8.1.2.1 DPR Architecture
The system architecture for the pattern recognition application directly
follows the DHGN clustered configuration described earlier. Figure 8.3 shows
the grid network layout for the proposed framework.
Communication between the DHGN subnets and the SI module is performed
using existing file transfer or resource allocation services, such as GridFTP
or GRAM. Each DHGN subnet can be hosted by a single computing node,
or group of nodes within a subnet. The communications between the nodes
in each subnet are handled by a Message Passing Interface (MPI), which
facilitates the parallel DHGN computations.
The proposed distributed pattern recognition is a real-time application that
is able to produce the results in a single cycle of computations. Furthermore,
each of the DHGN subnets executes independently, and thus provides a high
level of scalability and e
ciency by removing the requirement for inter-subnet
communications. The SI role and a DHGN subnet role can be easily inter-
changed, i.e., any node in the grid can take over the SI role for the framework.
