Databases Reference
In-Depth Information
that the use of facial images in the recognition simulation does not imply
that DHGN is a face recognition application with a promising high level of
accuracy. Rather, the simulation indicates the capability of DHGN to perform
distributed multi-feature recognition on complex patterns, such as gray-scale
images.
8.3 Distributed Data Management within Cloud Com-
puting
Existing data management and access schemes in clouds are mainly based
on Google File System (GFS) and MapReduce schemes. Problems arise when
data are partitioned between numerous available nodes therein. A new method
for partitioning and distributing data, known as resource virtualization in
cloud computing, has been explored by Basirat et al. [56]. Loosely coupled
associative computing techniques, which have not been commonly considered
for clouds, can provide the required breakthrough for data management in
Internet-scale infrastructures. Applications based on associative computing
models can e
ciently utilize the underlying hardware to dynamically scale
up and down the system resources. In doing so, the main hurdle to providing
scalable partitioning and distribution of data in clouds is removed, bringing
forth a vastly superior solution for virtualizing data intensive applications and
the system infrastructure to support the pay-per-use basis.
What is really required for any cloud system is a complete data access
scheme that enables data partitioning on-the-fly and has the ability to dissem-
inate processing nodes for specific data retrieval/storage tasks. A number of
possibilities have been explored to consolidate the data access scheme using an
e
cient partitioning approach. This integration within a complete end-to-end
scheme will enable data storage and retrieval processes to be performed effec-
tively, regardless of the distribution of data within the cloud system. The aim
is to develop a distributed data access scheme that enables data access to be
conducted effectively by means of the distributed pattern recognition (DPR)
approach. Data will be treated as a pattern, and data storage and retrieval will
be performed through an unsupervised pattern recognition mechanism. This
approach envisages data retrieval to be implemented as a distributed pattern
recognition process that is implemented through the integration of loosely
coupled computational networks. A divide-and-distribute approach allows for
the dynamic distribution of these networks within the cloud.
