An Energy-Efficient Framework for Dynamic Server Consolidation in Cloud Data Centers

Abstract

Cloud computing is a new internet-based computing which provides computing services as utilities, which are chargeable according to the usage and available on demand over the Internet. It allows the users and enterprises to store and process their data in a cloud infrastructure to let them focus better on their business needs. Rise in popularity of cloud computing has led to the growth of large scale data centers which need significantly higher level of energy consumption. As these data centers remains underutilized most of the time, a significant amount of energy get wasted in keeping the servers active. Efficient management of this energy consumption thus becomes a key issue to address. Various techniques are being developed these days for better power management in cloud. One such method is to dynamically adjust voltage and frequency of a processor according to its workload to reduce the energy consumption of underutilized servers. Another technique is VM consolidation which uses Live VM migration to pack the maximum number of VMs on minimum number of servers to maximize utilization, and thereby, decreasing energy consumption. In case of highly dynamic workloads, these techniques may lead to resource insufficiency in VMs when there is an increased load. Migration should be done in a way that there are no violations to Service Level Agreements (SLA) decided by the cloud consumers and service providers. So overall, there is a need of method that can maintain balance between energy efficiency and performance levels of a data center. Some methods have been proposed to handle this problem but very few of them considered the cost of migrating VMs in terms of time and energy. We have proposed a strategy for energy-efficient cloud data centers. It makes use of a prediction model based on historical data of workloads to anticipate the upcoming resource demands of applications. This enables identification of optimum number of servers required to host all the VMs, so that underutilized servers can be hibernated. This reduces the energy consumption in data centers without violating the required performance parameters. Our technique is migration cost-aware which means it takes into account the cost associated with the migration in terms of both energy and downtime. The simulation results show that our method provides a significant amount of energy conservation with minimal downtime and number of migrations.