@article{MAKHILLJEAS2019142418859,
    title = {Performance Evaluation of Static VM Consolidation Algorithms for
Cloud-based Data Centers with Predefined Machine Types},
    journal = {Journal of Engineering and Applied Sciences},
    volume = {14},
    number = {24},
    pages = {9810-9821},
    year = {2019},
    issn = {1816-949x},
    doi = {jeasci.2019.9810.9821},
    url = {https://makhillpublications.co/view-article.php?issn=1816-949x&doi=jeasci.2019.9810.9821},
    author = {Young-Chul},
    keywords = {Cloud-based data centers,energy-efficiency,inter-VM performance interference,SLA violations,static VM consolidation algorithm,resource requirement},
    abstract = {Energy efficiency in data centers is a very important issue and getting growing attention from
researchers. One approach to reduce energy consumption is to allocate tasks to Virtual Machines (VMs) created
in Physical Machines (PMs) in such a way that the number of idle PMs is maximized. Approaches of this kind
are called VM consolidation methods. Idle PMs can be put into an energy-saving sleep mode in which PMs
consume significantly lower energy than in the normal operation mode. But if too many VMs are packed into
a single PM, the performance interference among VMs can cause significant slowdown to jobs. When a new
job arrives at a cloud, the tasks of the job should be allocated to idle VMs. If there are enough number of idle
VMs, we should decide to which idle VMs those tasks should be assigned. If there are not enough idle VMs,
we should create necessary number of idle VMs on proper PMs before allocating the tasks to idle VMs. This
problem is called the static VM consolidation problem. In this study, we propose four algorithms for this static
VM consolidation problem. When we propose algorithms, we take following issues into considerations:
imperfect performance isolation of virtualization technology, flexible and efficient proactive VM creation policy,
PMs consisting of multiple CPUs each of which consists of multiple cores and VMs which are created
with pre-defined machine types. Further, we assume that we do not have the knowledge of the completion time
of a job, although, its resource requirements can be known a priori. We analyze the proposed algorithms
through simulation with synthetic workloads obtained by analyzing the characteristics of workloads in real data
centers. We measure following three metrics and suggest the best algorithm: ratio of idle PMs, service level
agreement violation ratio and the total energy consumption in a cloud.}
    }