Information Technology Reference
In-Depth Information
k VM is denoted as VM Application Sequence , where Application Sequence
is the sequence of applications ID whose instances are running in VM . For example,
if 1 instance of application 1, 1 instance of application 2 and 1 instance of application 3
are running in k VM, this VM can be denoted as VM 1,2,3 . Application Se-
quence probably has duplicate application ID since there could be more than one in-
stances of one application running simultaneously on the same VM. The above objec-
tive is achieved by introducing the heuristic algorithm and removed the instances of the
VM which has the least number of running instances and its instances are the subset of
the instances to be removed. Assuming there are 4 VMs: VM 1,2 , VM 1 ,
VM 2,2,3 , VM 3,4,5 , and the instances to be shrunk is {1,1, 2, 2, 2, 3}. Based on
our scale-down algorithm, the instances should be removed according to the following
sequence: VM 1 , VM 1,2 and VM 2,2,3 . VM 1, 2 and 3 are set idle after the
scale-down and can be returned to IaaS Layer.
4.4
Resource Management Process
In summary, the resource management process in PoI is essentially a feedback-control
loop. At the beginning of every period, Platform Master collects all feedback data
from other components of the platform and makes resource management decisions
according to the feedback data. Every feedback-control loop can be divided into 4
phases.
To collect performance metrics including Missed Deadline Ratio and CPU Utiliza-
tion of every application deployed on the platform.
To compute ∆Instance for every application based on PID control function men-
tioned above.
To deploy or remove application instances based on scale-up or scale-down algo-
rithms. Rent VMs from IaaS Layer if needed.
To return all idle VMs to IaaS Layer in order to save rental cost.
5
Experiments
In this chapter, the comprehensive evaluation of the resource management framework
of PoI is presented. In Section 5.1, the details of the experimental setup are specified.
In Section 5.2, the proposed framework under different experimental parameters is
illustrated, then the analysis of the experimental results is given.
5.1
Experimental Setup
For the evaluation of the proposed resource management framework in PoI, work-
loads of different patterns as the workloads of the platform are prepared. The different
workloads are provided by The Grid Workloads Archive, which could provide ano-
nymous workload traces from grid environments to researchers and to practitioners
alike [13]. 3 workloads of different patterns within 24 hours are chosen as shown in
Figure 5.
Search WWH ::




Custom Search