Search

Server consolidation in a virtualized cloud environment improves resource utilization by hosting many virtual machines (VMs) on a single machine. The host server failure induces a serious risk of the interruption of services provided through VMs hosted on the failed server. Ensuring high reliability and availability from the unpredicted host server failures is a challenging task, and this assurance guarantee is the quality of service to the users. Executing latency-sensitive workloads within a time frame, multireplica strategy-based reliability assurance is used in general. In this article, we propose two reliability-ensured scheduling approaches by strategically allocating the jobs to machines to satisfy their reliability and deadline requirements. The proposed approaches “Replication with equal failure rate” and “Efficient reliability replication method” are used to schedule the jobs to machines with equal and different failure rates, respectively, while satisfying both the deadline and reliability requirements of the jobs. Both the scheduling approaches dynamically select the machines for the jobs to minimize the number of replicas based on job characteristics and failure rate of machines. Also, we adopt an overlapping mechanism to further minimize the resource utilization. The extensive simulation results on randomly generated and real-world data at different scales show that the proposed approaches perform better than other state-of-the-art approaches. [ABSTRACT FROM AUTHOR]

• Nonlinear relationship between power consumption and utilization of the cloud hosts. • Urgent points for real-time tasks in a heterogeneous computing environment. • Two energy efficient scheduling policies for online aperiodic real-time tasks. • Extensive simulation work on CloudSim toolkit for synthetic and real-world data Cloud computing platform has emerged to be a promising computing paradigm of recent time. Various applications from different domains having rigid deadline constraints are deployed in the cloud system for their respective benefits. Energy-efficient execution of these applications, meeting their deadline constraints is a challenge. Most of the existing research on the energy-efficient scheduling of these applications in the cloud domain consider a linear relationship between the energy consumption and the resource utilization of the system, and they focus on maximizing the utilization of resources to reduce the active number of computing nodes to minimize energy consumption. In this paper, we first devise a power consumption model for the cloud system which considers both the static and dynamic components of it and assumes a nonlinear relationship with utilization. Then we introduce the concept of urgent points in case of tasks having deadline in the context of a heterogeneous cloud computing environment. Then we propose two energy-efficient scheduling approaches, named UPS and UPS-ES designed based on the urgent points of the tasks and two threshold values of the host utilization. Extensive simulation experiments are conducted both for synthetic tasksets and Google cloud tracelogs. The results are compared with a state of the art scheduling policy and found that our policies perform significantly better than them, with an energy reduction of up to 42% while the deadline constraints of all the tasks are met. [ABSTRACT FROM AUTHOR]