Information Technology Reference
In-Depth Information
Accordingly, the aim of this research is to develop a multi-objective mechanism to
determine the lowest cost for virtual machine (VM) placement that considers various
components of cost efficiency (such as electrical power consumption, availability, and
load balancing).
Traditional systems have been designed and built to satisfy the peak load that was
estimated in advance. However, although periodic fluctuations for business workloads
may be predictable, it is difficult to predict fluctuation of demand for a public cloud.
Load balancing among physical servers is a typical criterion to increase the efficiency
in cloud computing data centers. Therefore, many existing resource schedulers
determine VM placement on the basis of utilization of physical servers (typically CPU
and memory usage) or VM hosts. However, efficient and effective VM placement
involves many other factors, and existing resource schedulers do not give these
additional factors adequate consideration because multi-objective optimization is an
extremely complex problem. If this problem is addressed in a straightforward way, an
astronomical number of possible combinations will be generated because of the large
number of physical servers and VMs in a cloud computing data center; i.e., a huge
number of time-consuming computations are required to obtain an optimum result.
Consequently, timely optimization is a challenge. A VM deployment request from
a user should be completed within several minutes (including time to boot up VM).
Therefore, a very limited amount of time is available to determine an optimum VM
placement.
2
Various Constraints to Virtual Machine Placement
In addition to the physical capacity of each host (i.e., CPU and memory), resource
scheduling often has to consider various other constraints. These constraints are
related to policies established by the operator of the cloud. Moreover, it is common
for the constraints to conflict with each other. Examples of constraint policies are
given below:
Efficient use of electricity (towards a green environment): This policy allocates
as many VMs as possible to a host. Electricity can be saved by powering-off
idling hosts.
Availability of a virtual system: This policy distributes VMs to different hosts
(redundancy) to guard against VMs going down if a single host fails. This policy
may conflict with the goal of saving electricity.
Affinity: This policy allocates compatible or similar VMs to the same host. For
example, network traffic via switches and routers can be reduced by allocating
VMs that routinely communicate with each other to the same host. This is a
typical situation for multiple VMs owned by a single tenant in a multi-tenant
data center.
Repulsion: VMs that compete for resources should not be placed on the same
host. For example, VMs requiring higher network bandwidth should be placed
repulsively to avoid network congestion. Firewalls are a typical example.
