Energy-Aware Policies in Ubiquitous Computing Facilities - Cloud Computing with e-Science Applications - page 271

reduce the computational power of servers by means of energy-efficient
scheduling techniques [12, 13], resource allocation, and workload assignment
mechanisms [14, 15]. For cloud computing applications, virtualization tech-
nology has provided a promising way to manage application performance
by dynamically reallocating resources to virtual machines (VMs). Several
management algorithms have been proposed to control the application per-
formance for virtualized servers [16] and to solve the VM-server mapping
problem for power savings [17].
In our proposal, scheduling and resource allocation take into account the
global energy consumption, which includes the cooling consumption of the
data center and the consumption of the rest of the system: node and PDA or
smartphone. So, it is possible to take advantage of the heterogeneity of the
system and download only some part of the computation to the data center,
while the rest is performed in the PDA.
12.3 Proposed Novel Paradigm
12.3.1 Devised Computer Paradigm
Next-generation applications are usually composed of a large number of
sensors, wirelessly connected to the cloud through a mobile processing
device. Data centers provide cloud-based data services that can closely match
the demand of processing capacity, according to data size and complexity of
the analysis algorithms. By sharing data center resources for multiple appli-
cations, it is possible to reduce the need for resources, maintaining high utili-
zation rates and reducing energy requirements. To provide adequate energy
management, this heterogeneous distributed computing system is tightly
coupled with an energy analysis and optimization system, which continu-
ously adapts the amount of processing that is performed in the different
layers of the distributed system and the resources assigned to each task.
12.3.2 Energy Optimization System
Figure 12.1 shows the proposed system architecture for the energy optimi-
zation of cloud computing in e-science applications. Detailed functions of
constituents in the system are summarized as follows:
• Application support network: Applications require a heterogeneous
network comprising sensor nodes, data centers, and some kind of inter-
connection network to drive data from sensors to data centers. Each
element has different computation capacity, functional requirements,
power consumption characteristics, and so on.