Energy-Aware Policies in Ubiquitous Computing Facilities - Cloud Computing with e-Science Applications - page 281

number of resources and tasks to allocate, GAs behave much better in terms
of performance.
One of the benefits of using a GA is the possibility of tackling a large set
of constraints (the maximum temperature of the servers, the available CPU
capacity, the required instructions per task, etc.). In this way, the GA defines
a vector of n decision variables, a vector of m objectives function, a number
of constraints not satisfied, the total energy, and the feasible region in the
decision space. The algorithm allows unfeasible solutions, but only when no
other alternatives are found.
For the chromosome encoding, each gene represents a decision variable.
Because many decision variables are integers, the chromosome uses integer
encoding. Thus, some decision variables (like the CPU capacity) are scaled to
the integer interval and transformed to a percentage when used in the multi-
objective function for evaluation. The evolutionary solver starts with a random
population of chromosomes. After that the algorithm involves the population
applying (1) the non-dominated sorting genetic algorithm (NSGA-II) standard
tournament operator, (2) a single-point crossover operator with probability
of 0.9, (3) an integer flip mutation operator, and (4) the multiobjective evalua-
tion. Steps 1 to 4 are applied for a variable number of iterations or generations.
Using this approach, it is possible to obtain optimal energy savings, realistic
with the current technology, in much shorter time than traditional algorithms
and targeting much more complex environments.
12.6 Resource Selection and Configuration
Cloud computing presents a compelling opportunity to reduce data center
power bills. The economic advantages of shifting to a cloud infrastruc-
ture are enormous, and current challenges in cloud adoption will be over-
come soon, leading a major shift to cloud computing. In this computational
context, the goal of techniques like “resource selection” and “configuration”
is to offer new services more efficiently by properly selecting and configur-
ing the available resources. The algorithms described in the previous section
can be jointly applied with the cloud-specific techniques proposed in this
section—virtualization, consolidation, and managing the operating server
set—to substantially increase energy savings.
12.6.1 Virtualization
Virtualization allows the management of the data center as a pool of
resources, providing live migration and dynamic load balancing, as well as
the fast incorporation of new resources and power consumption savings.
In addition, a single node can accommodate simultaneously various VMs