Energy-Aware Policies in Ubiquitous Computing Facilities - Cloud Computing with e-Science Applications - page 283

Moreover, the cost of the machine turned on or off to suit the operational
farm-to-user demand also must be taken into account. This cost involves two
important factors:
• Energy: Consumption of machines when turned off and on again is
significant [40]. The energy saved during the period in which servers
are switched off should be compensated by this offset energetic cost.
• Delay: The server turn on takes a certain time, so the incoming
demand and its variations have to be anticipated. Backup physical
machines should be available to host peak requirements.
Currently, one common technique is to apply low-power modes to inactive
servers to save static energy [41]. This policy helps minimize delays when
activating new machines under peak demand, reducing consumption of idle
servers. Many servers offer sleep or hibernate states, such as standby modes,
that consume less than active modes with different setup times. Finally, it is
necessary to take into account these additional costs in resource configura-
tion policies to minimize energy globally.
This technique can be combined with dynamic voltage and frequency
scaling (DVFS). Dynamic consumption can also be reduced by acting on the
low-power modes of the machines at runtime, but only if this policy does
not violate QoS requirements contracted by users. Modifying the frequency,
voltage, or both varies the response time, affecting the completion of services
and applications. Decreasing the frequency or operating voltage reduces
dynamic power consumption during the execution of a workload. Also,
during idle periods, the static consumption is minimized at low voltages
and frequencies.
Therefore, if QoS restrictions are not strict, energy savings in the computing
part can be increased by the efficient application of the presented techniques.
12.7 Conclusions
Cloud computing, MCC, or even modern HPC start with data centers. While
we can dream of a world in which anyone is allowed to sell their excess
computing capacity as virtualized resources to anyone else or where the
ubiquitous sensing of information is processed by a center kilometers away
from the source, the fact of the matter is that today the cloud finds strong
energy constraints because of the energy-hungry computing “factories.”
However, data centers are not the only computing resources that contrib-
ute to the energy inefficiency. Distributed computing devices and wireless
communication layers are also responsible for this.