Power-Aware Allocation of Virtual Machine-Based Real-Time Cloudlets in Cloud Data Centers

Due to the expanding utilization of cloud
computing services, power consumption in cloud data centers
has increased significantly. The number of active physical hosts
impacts data center power usage, so the number of active
physical hosts should be decreased. To achieve this goal, cloud
data centers use virtualization technology to consolidate
multiple virtual machines on a single physical server, using
state-of-the-art virtual machine placement algorithms.
Specifically, bin packing algorithms have been widely used to
place a set of items, i.e., cloudlets and virtual machines, into a
set of bins, i.e., virtual machines and physical hosts. However, a
set of cloud services, i.e., cloudlets, are characterized as realtime and need to be provided within strict deadlines. In this
paper, a cloud resource allocation framework is proposed to
provide a compromise between two goals. The proposed
framework uses the optimal physical host MIPS to achieve
minimum possible power consumption while satisfying virtual
machine-based cloudlets' deadline constraints. The proposed
framework includes two modules, namely cloudlet allocator and
virtual machine allocator. A set of widely used bin packing
algorithms is exploited and compared in both modules. Firstly,
the algorithms exploited in the cloudlet allocator module include
first-fit, best-fit, and round-robin. The evaluation results
showed that the round-robin algorithm provides the best
outcomes in terms of real-time constraints. Round-robin could
allocate an increasing number of cloudlets to virtual machines
without scarifying the deadline constraints. Secondly, the
algorithms used in the comparison in the virtual machine
allocator module include first-fit, best-fit, next-fit, and worst-fit.
The results showed that the best-fit algorithm reduces power
consumption among all other algorithms under consideration.
The results also suggest that setting the physical host CPU MIPS
to optimal MIPS achieves the least consumed power.