Information Technology Reference
In-Depth Information
Several indicators have been introduced to measure Data Center eciency
under the vision of achieving economical, environmental and technological sus-
tainability [13]. In this context, Green Performance Indicators (GPI) are defined
as the driving policies for data collection and analysis related do energy con-
sumption. The idea of GPIs is interesting because it can be adapted as part
of Service Level Agreements (SLA), where requirements about energy eciency
versus the expected quality of services are specified and need to be satisfied.
The GPIs are classified in four clusters (IT Resource Usage, Application Lifecy-
cle, Energy Impact and Organizational). In this work, we consider only the IT
Resource Usage and the Energy Impact GPI clusters.
The IT Resource Usage GPIs characterize the energy consumption of an ap-
plication as a function of the energy consumed by its resources. Examples of
metrics are
CPU usage
,
Memory usage
and
I/O activity
.
The Energy Impact GPIs describe the impact of Data Centers and applica-
tions on the environment, considering power supply, consumed materials, emis-
sions, and other energy related factors. The most important Energy Impact GPI
metrics are: a) application performance indicators, which measure the
energy
consumption per computing unit
, using typically FLOPS/kWh or Number of
Transactions/kWh; b)
Data Center Infrastructure Eciency (DCiE)
,whichis
used to determine the energy eciency of a Data Center as a whole; and c)
Compute Power Eciency (CPE)
, which computes the data center power. In
this metric, the power consumed by idle servers is computed as overhead.
4 Design of the Multi-Agent Consolidation Mechanism
The main goal of our approach is to meet the QoS requirements of the applica-
tions, while keeping the power consumption of the Data Centers below a given
energy threshold defined by a third party agent. To achieve this goal, we propose
a Multi-Agent strategy to negotiate the resource allocations among Clouds.
We consider a federated Cloud environment with four distinct agents: Cloud
Service Provider (CSP), Cloud User (CLU), Energy Power Provider (EPP) and
Carbon Emissions Agency Regulator (CEAR) as shown in Figure 4(a). In our
design, the CEAR determines the amount of carbon emissions that both the
CSP and the EPP can emit in a period of time.
In each Data Center, there is one coordinator responsible for monitoring data
center metrics, negotiating with the other agents. There are also sensors to mon-
itor energy consumption, resource usage and SLA violation as shown in Figure
4(b). The scenario proposed is a set of Data Centers composed of a set of Virtual
Machines, which are mapped to a set of physical servers that are interconnected
anddeployedinahybridcloudmodel.
Let
R {r
1
,r
2
, ··· ,r
n
}
be the set of resources in Data Center
i
with a capacity
c
i
,where
k ∈ R
. The Energy Consumption for the Data Center (
E
i
) is defined
as
E
i
=(
p
max
− p
min
)
∗ U
i
+
p
min
[14], where
p
max
is the power consumption
at the peak load,
p
min
is the minimum power consumption in active mode, and
U
is the resource utilization of Data Center
i
as defined in
U
i
=
j
=1
u
i,j
[14],
where
u
i,j
is the resource usage of resource
j
in Data Center
i
.
