Local Regression Based Box-Cox Transformations for Resource Management in Cloud Networks

摘要

Understanding and implementing approaches to efficiently manage the infrastructure resources of cloud data centers has become essential. Energy consumption and disorganized resource usage can expensively produce an impressive increase in the operational cost of cloud services. This increase turns to a remarkable rise in the cloud customers' invoices. Providing an exceptional quality of service running on well-organized resources with efficient energy is a critical issue that needs to be carefully considered by both industrial and academics. Although, the cloud providers are trying to deliver sufficient quality of services to their customers with a comparatively proper cost. One of the effective techniques to address these issues in cloud data centers is a dynamic virtual machine consolidation. This technique intends to improve energy efficiency and resource utilization by reallocating multiple virtual machines including various workload among available hosts and turning the unutilized hosts to an ideal state. However, consolidating the virtual machines due to fluctuating workload in cloud application can cause a violation in service level agreement. In this paper, we propose a host overload detection algorithm based on the Box-Cox transformations and the local regression model to predict overloaded hosts. This algorithm transforms the historical data of the host workload by using the Box-Cox transformations technique, and it also applies the local regression to predict the future state of the selected host. The experiments and simulation results based on dynamic workloads show the proposed algorithm outperforms the other competitive host overload detection algorithms.