Information Technology Reference
In-Depth Information
We are proposing an algorithm, which will be described in detail in a following
section, which utilizes a representative VM from an equivalent set to minimize the
number of computations. First, because we need to check the status of each host, for
each constraint, the order of computations to classify
n
hosts into equivalent sets is
O
(
n
). It is also
O
(
n
) for
k
constraints. Note that it is reasonable to assume
k
<<
n
.
Second,
O
(
n
) computations are required to calculate a comprehensive cost for
n
hosts.
Additionally, if all the hosts are neighboring, a maximum of
O
(
n
) computations are
needed to conduct a neighborhood search. All these computations of
O
(
n
) are
necessary for
m
VMs. Therefore, the order of computation is
O
(
n
*
m
).
For example, for 120 hosts and 8 VMs:
O(
n
m
) = 120^8 (~ 10^16): brute-force method
O(
n
*
m
) = 120*8 (~ 10^3): proposed mechanism.
{
Flexibility and Arbitration.
Multi-objective optimization must be flexible and
capable of arbitrating conflicting constraints. For example, electricity saving and
availability of a virtual system are compatible policies in a multi-tenant environment.
Although the latter policy acts to distribute one tenant's VMs on different hosts, the
former policy works to cluster several tenants' VMs on one of host. As a result, the
number of powered-on hosts can be decreased by sharing hosts among tenants.
Considering other optimization objectives, the proposed mechanism
programmatically finds an optimum VM placement from a large number of possible
combinations.
The structure of the proposed mechanism is illustrated in Figure 1. It consists of
following three modules:
1.
Cost Evaluation Plug-in Module
The proposed mechanism has been designed with a plug-in structure to enable
data center operators to implement various operational policies. The cost
evaluation plug-in module evaluates the cost on the basis of a specific
optimization objective function. The exceptional value of this design feature is
that is allows a comparison of the impact of various constrains on the basis of
the cost.
Fig. 1.
Structure of the proposed mechanism
