HCloud, a Healthcare-Oriented Cloud System with Improved Efficiency in Biomedical Data Processing - Cloud Computing with e-Science Applications - page 167

files to access metadata on the local file system, which may result in syn-
chronous problems (e.g., on the GlusterFS [13]) . So, physiological signals with
various formats should be well managed and processed to provide efficient
instant services to individuals, and a consistent storage system is needed to
adapt to this situation.
8.2.1.2 Multiscale File Storage and Integration
There are mainly two categories of application data in the system. One is
trivial and has a small amount of data for temporal signals, such as the afore-
mentioned physiological signal data; the other has huge-scale graphic data
generated by drawing server clusters. Those large numbers of semistruc-
tured or unstructured health records, as well as massive trivial files, are all
adapted to NoSQL [14] (not only in SQL) databases instead of traditional rela-
tional ones, which has natural advantages for easily expanding horizontally.
The key idea of NoSQL is that it employs a loosely coupled data model and
has neither a fixed table schema nor joint operations. Hence, it is appropriate
for the high-performance requirement when accessing large files, especially
for those without fixed structure.
8.2.1.3 Adaptive Algorithms for Different Targets
In consideration of the various characteristics of body health and aspects
of a monitoring system, we need different algorithms for different signal
processing and data mining. Note that the analysis routines should be easily
configurable and adaptive to concurrent requests from users. Therefore, a
flexible algorithm scheme should be developed for the sake of on-demand
services. To cope with the irregularity of the data structure, a self-defined
message head would be utilized to identify the call of various routines.
Moreover, many real-time tasks should be addressed, and high concurrent
mechanisms will be the major concern, although a general cloud may not
need to face too many real-time transactions.
8.2.1.4 Visualization of Health Analysis
Another challenging task is the visualization of analysis results, which are
usually computationally intensive with a large amount of graphic data for
drawing. Careful considerations should be taken into account for efficient
storage of and access to the huge graphic data generated by the analysis
results. The system should be capable of handling the different types of data
visualization adaptively. These fundamental characteristics are very differ-
ent from the features of grid computing, which aimed at special applica-
tions and were difficult to operate for unprofessional users (e.g., in scientific
exploration projects).