Information Technology Reference
In-Depth Information
Minimum migration cost: This policy determines VMs that should be migrated
from one host to another host. Migration costs can be a function of factors
memory size, I/O rate, and the number of hops between hosts.
3
Challenges
Realization of Practical Response Time.
It is not practical to use a brute-force
method to find optimum VM placement because of the large numbers of hosts and
VMs in a cloud computing data center. To realize practical computation time, the
number of possible combinations must be limited.
Arbitration.
As mentioned in the previous section, the various policies established by
the cloud computing data center to determine VM placement may conflict with each
other; for example, reducing power consumption conflicts with high availability, and
conversely, ensuring high availability may increase power consumption. One of the
challenges for a cloud administrator is determining how to arbitrate such conflicting
constraints. Prioritization could be a solution, but arbitrating the conflict between
availability and power consumption is not axiomatic.
Flexibility.
Since the requirements and prioritization of policies can differ among
data centers, a multi-objective optimization system should allow a data center
operator to configure (add and remove) policies to satisfy particular requirements.
Moreover, any system for determining VM placement must be sufficiently flexible to
respond to changes in economic, political, and social conditions, such as the rising
cost of energy, preferential taxation systems for ecological initiatives, regulations
requiring the restriction of CO
2
emissions, and other responses to global warming and
climate change. For example, data centers in Japan had to respond to restricted
electricity supply after the earthquake and tsunami, which occurred in March, 2011.
4
Solutions in the Design
Avoiding Excessive Numbers of Possible Placement Combinations.
This is
essential for the realization of a practical multi-objective optimization mechanism.
We have solved this difficulty by defining and introducing equivalent sets of hosts.
Although a large number of hosts exist in a cloud computing data center, they can
be classified into a relatively small number of equivalent sets on the basis of the status
of each host. One host from a set can be used to evaluate the cost, thereby
significantly reducing the required computations. For example, all hosts can be
classified into two equivalent sets (powered on and off) to determine how to apply the
electricity-saving policy mentioned in section 2.
Here, we assume that
n
is the number of hosts and
m
is the number of VMs to be
deployed. When a brute-force method is used, there are
n
choices for placement of
each VM. Therefore, the order of required computation is
O
(
n
m
).
