Information Technology Reference
In-Depth Information
6. If all negotiations fail, the Cloud Provider finds the SLA whose violation
implies in lower cost and execute the task. In this case, the price to execute
the tasks is defined as shown in Equation (2).
V
t
=
P
t
+
γ
+
δ
(2)
where
γ
is the cost of violate the QoS requirements of other tasks and
δ
is
the cost associated with energy consumption violation.
To control task allocation, each Cloud Provider has a matrix representing
tasks
t
i
∈ T
, virtual machines
vm
j
∈ VM
and physical servers
r
z
∈ R
,where:1
represents that
t
i
is allocated at
vm
j
in resource
r
z
; 0 indicates that
t
i
can be
allocated at
vm
j
; and -1 represents that this allocation is impossible.
In order to illustrate our strategy, consider a federated Cloud with 2 Data
Centers (DC1 and DC2) and a user that contracts DC1 to execute his applica-
tions. Consider that DC1 is overloaded and that the QoS requirement described
in the SLAs is response time. In this scenario, when the user submits tasks to
execute, the DC1 Cloud Provider first tries to execute them locally, considering
energy consumption and the available resources. Since DC1 is overloaded, its
Cloud Provider contacts DC2 and negotiates with it the execution of the tasks.
If DC2 accepts it, the cost of the tasks execution is calculated with Equation (1).
If DC2 does not accept, then DC1 tries to consolidate its virtual machines and,
if not possible, it tries to negotiate the energy threshold with the CEAR and the
EPP agents. If all negotiations fail, then DC1 finds the SLA whose violations
implies in lower cost and terminates the execution of its associated task. In this
case, the cost to execute the tasks is calculated with Equation (2).
5 Experimental Results
In this section, we present the evaluation of the strategy proposed in Section 4.
We used CloudSim [3], which is a well-established Cloud simulator that has been
used in many previous works [16], [18], among others. It is a simulation toolkit
that enables modeling and simulation of Clouds and application provisioning
environments, with support to Data Centers, Virtual Machines and resource
provisioning policies.
Since we are dealing with federated environments, we extended CloudSim by
adding four classes (
CloudEnergyReg, DCEnergySensor, FedPowerVMAllocPol-
icy, CustomerDCBroker
) to it, as well as isolation of queue events and support
for concurrent execution.
The
CloudEnergyReg
class represents the behavior of the CEAR agent. This
agent communicates with the Data Center cloud coordinator to inform the en-
ergy consumption threshold. The
DCEnergySensor
class implements the Sensor
interface that monitors the energy consumption of the Data Center and informs
the coordinator. When the energy consumption is close to the limit, this sensor
creates an event and notifies the coordinator, that can take actions. The
Fed-
PowerVmAllocPolicy
class extends the
VmAllocPolicy
class to implement the
proposed federated server consolidation mechanism Virtual Machine allocation
