Information Technology Reference
In-Depth Information
scale-up and scale-down which can lower the Time Usage of VMs as much as possi-
ble should be determined. The VM Scheduling Policy can be divided into two algo-
rithms: scale-up algorithm and scale-down algorithm.
1.
Scale-Up Algorithm
In order to decrease the Time Usage of VMs, the CPU capacity of every VM should
be monitored. In addition, there is a threshold for each VM to limit its CPU utilization
in a safe level (e.g., 90%). The objective of scale-up algorithm is to make sure all
instances are converging to a small number of VMs as well as keep each VM's CPU
utilization in a safe level. A memory management algorithm “BEST FIT” is intro-
duced to achieve the above objective. BEST FIT is a famous algorithm for OS memo-
ry management in order to avoid using larger blocks unnecessarily. In BEST FIT, the
block list is searched for the block that is smallest but greater than or equal to the
request size [12].
The suitable VM, into which new instances of the particular application to be ex-
panded are deployed, is chosen based on BEST FIT. Next this algorithm will be fur-
ther illustrated with an example. Assuming the utilization of all VMs is as figure 4:
Fig. 3.
CPU Utilization of VMs
In figure 4, there are 4 VMs. White box represents the percentage of CPU unused.
In this example, an instance of application j need to be created and this instance will
occupy 10% CPU utilization of a single VM. The safe threshold of VMs is 90%.
Therefore, the actual percentages of CPU which could be used are 23%, 10%, 35%
and 2% respectively. According to the BEST FIT algorithm, VM2 is chosen as the
target to bear application j's instance. If there is more than one instance to be scaled
up, the above procedures should be performed iteratively based on BEST FIT algo-
rithm. By using BEST FIT algorithm, the CPU of every rented VMs will be fully
exploited over the time. Scale-Down Algorithm
The scale-down algorithm is designed following the same objective as the scale-up
algorithm. Differing from scale-up algorithm, the scale-down algorithm tries to deter-
mine which instance to be removed in order to acquire as many idle VMs as possible
and return them to IaaS Layer.
