Energy-Aware Policies in Ubiquitous Computing Facilities - Cloud Computing with e-Science Applications - page 278

range of ambient temperatures, the impact of the temperature-dependent
leakage is negligible, whereas for a higher-temperature range leakage needs
to be considered.
To ensure the reliability of the IT equipment, CPU temperatures should
not increase above a certain threshold. The ASHRAE (American Society of
Heating, Refrigerating, and Air-Conditioning Engineers) [29] organization
publishes metrics on the maximum inlet air temperature for a server, the red-
line temperature, as well as the appropriate temperature and humidity con-
ditions of the data room environment to ensure that reliability is not affected.
Data room modeling is still an open issue, as the only feasible ways to
model the thermal behavior of the data room and be able to predict the inlet
temperature of the servers is either by deploying temperature sensors in the
data room that take measurements or by performing time-consuming and
expensive computational fluid dynamics (CFD) simulations. CFD simula-
tions use numerical methods to analyze the data room and model its behav-
ior. However, these simulations do not often match the real environments
and must be rerun every time the data center topology changes.
12.5 Ubiquitous Green Allocation Algorithms
Resource management is a well-known concept in the data center world and
is used to allocate in a spatiotemporal way the workload to be executed in
the data center, optimizing a particular goal. Traditionally, these techniques
have focused on maximizing performance by assigning tasks to computa-
tional resources in the most efficient way. However, the increasing energy
demand of data center facilities has shifted the optimization goals toward
maximizing energy efficiency. Works proposing allocation algorithms have
traditionally applied greedy algorithms [30], Markov chain algorithms [31],
mixed-integer linear programming (MILP), or mixed-integer nonlinear pro-
gramming (MINLP) [32] to generate the best task allocation. Most of these
approaches do not propose a precise objective function or accurate math-
ematical formulation of the optimization problem. Although some of these
solutions behave well in homogeneous data-center-level scenarios, they do
not consider the heterogeneity inherent in smart environment applications.
Moreover, MILP solutions do not scale well for larger scenarios with a high
number of servers and large workloads to allocate.
Only very recently industry and research started to agree on the impor-
tance of environmental room monitoring [33] to improve energy efficiency.
Other research [34] presented the data center as a distributed cyberphysi-
cal system (CPS) in which both computational and physical parameters
can be measured with the goal of minimizing energy consumption from a