Information Technology Reference
In-Depth Information
K
|P
k
− R
k
|
max(
R
k
)
K
K
(1)
|T
ref
− T
k
|
(2)
max(0
,FT
(
i
)
− D
(
i
))
(3)
k
=1
i
=1
k
=1
+
S
idl
k
+
S
sleep
and
S
sleep
k
The total IT power is
I
k
=
S
max
where
S
max
k
,
S
idle
k
k
k
are the total power of servers executing, idle, and sleep at time
k
.
AC is used to keep temperature within a specified hysteresis band. In free
cooling mode, outside air is blown into the datacenter using a fan. The cooling
power
C
k
takes the values
C
PWR
in AC mode, compressor ON; 0 in AC mode,
compressor OFF; and
FAN
PWR
in free cooling mode. Based on two months of
operation data from Parasol, we identified an Auto-Regressive eXogenous [18]
model for temperature:
A
(
q
)
T
k
=
B
(
q
)
u
k
+
e
k
.
4 A Multiobjective Evolutionary Approach for
Energy-Aware Datacenter Planning
This section presents the proposed MOEA for energy-aware datacenter planning.
4.1 Evolutionary Algorithms and NSGA-II
Evolutionary algorithms (EAs) are non-deterministic methods that emulate the
natural evolution to solve optimization problems [3]. In the last thirty years,
EAs have been successfully applied for solving many optimization problems.
MOEAs [5] have been applied to solve hard optimization problems, obtaining
accurate results when solving real-life problems in many research areas. Unlike
many traditional methods for multiobjective optimization, MOEAs are able to
find a set with several solutions in a single execution, since they work with a
population of tentative solutions.
MOEAs must be designed aiming at two goals at the same time:
i
) approxi-
mating the Pareto front, by applying a Pareto-based search, and
ii
) maintaining
diversity instead of converging to a reduced section of the Pareto front, by using
specific techniques from multimodal optimization (sharing, crowding, etc.).
In this work, we apply the
Non-dominated Sorting Genetic Algorithm, version
II
(NSGA-II) [5], a popular MOEA that has been successfully applied in many
application areas. NSGA-II includes features to deal with specific issues of the
search:
i
) a non-dominated elitist ordering that diminishes the complexity of the
dominance check;
ii
) a crowding technique for diversity preservation; and
iii
)a
fitness assignment method considering the crowding distance values.
4.2 The Proposed Resolution Approach
Solution encoding
: Each solution represents the power (cooling and IT) to be
used at each time step
k
. A solution is encoded as an integer vector of 2
K
elements, representing the cooling (positions 1 to
K
) and server power (positions
K
+1 to 2
K
). The server power is encoded directly as Watts, whereas the cooling
power is encoded as an integer value representing three states: (a)
1-100
:free
