Energy-Efficient Task Scheduling Techniques And Applications In Cloud

With the development and applications of information technology,the volume of data generated in the applications grows explosively.Traditional database management system cannot deal with the massive data,high concurrency,rapid response and scalability required by big data.How to store and manage the big data efficiently is urgent.Cloud computing with characteristics of low price,scalability and fault tolerance has become one of the key technologies of big data management.As cloud computing has been widely used in many applications,the number and the scale of the data centers increase raplidly,and the energy cost has exceeded the acquisition cost of the hardware devices.The increasing energy demand of data centers challenges the energy crisis and environment protection in the world.Therefore,energy efficient data management in the cloud is essential and enough attentions should be paid to it.Energy efficient task scheduling,which aims to reduce the energy consumed by the nodes running the tasks,is one of the most important parts of energy efficient data management in cloud computing.In this paper,energy efficient task scheduling in the cloud is studied from the aspects of single node and the cluster of multiple nodes.The innovations of this paper are given as follows:(1)The existing methods of energy efficient task scheduling on multicore server typically assume that the cores can be controlled independently and the power of components of the server except the processor is not taken into account,and the energy efficiency defined in the methods often makes no sense.To overcome the issues,we propose a uniform framework for energy efficient task scheduling on multicore server.The static and dynamic power of the processor and the power of other components of the server are taken into account,and the economic cost of the energy efficiency is defined as the weighted sum of processing time and the energy consumption.Different scheduling algorithms are developed for the processors with various core architectures.The experimental results show that more cost will be reduced when the load becomes heavier if the cores are independentlly controlled on the sever,the cost of our proposed method is less than the traditional methods and the differnece between them becomes larger when processing heavier load for the chip controlled processor,and if the processor is island controlled,the cost will be reduced exponentially compared to the existing methods.(2)Since the methods of energy efficient data-intensive task scheduling in the cloud modify the data placement policies,they cannot suit for all applications.An energy efficient data-intensive task scheduling algorithm EABD,which is independent to the data placement strategies in the cloud,is proposed in this paper to reduce the energy consumption for task execution.Though more nodes will be used to process the task,EABD consumes less energy,and in some cases over 50% energy reduction will be made.Meanwhile,the number of replications has little affect on the energy consuption of EABD,and the least energy will be consumed when each data chunk has 3 replications.(3)To deal with the heterogeneity of the nodes in practical cloud,MinBalance,which is an algorithm of energy efficient task scheduling in cloud with heterogenous nodes,is proposed based on EABD.The process is devided into two steps,node selection and workload balance.In the process of node selection,four definitions are introduced to weight each node and the node with least weight will be selected to process the task in each iteration.The workload of the selected nodes will be balanced to reduce the energy waste caused by waiting in the process of workload balance.The heterogeneity of the performance and power of each node are taken into consideration to reduce the energy consumption,MinBalance may lead to nearly 60% energy reduction if the data set is very large.(4)Using minimal nodes to process the virtual machined is the main idea of exsiting methods of energy efficient scheduling of virtual machines in the cloud.A novel energy efficient algorithm of virtual machines scheduling EEVS is proposed.The VMs are allocated to minimal nodes with high ratio of performance to power to reduce the nergy consumption in the node layer.When the VMs are processed,the DVFS-based technology is adopted to reduce the energy consumed by each physical machine in the layer of hardware components by VM migrations and resource consolidations.EEVS may lead to over 10% energy reduction without performance degradation.(5)Cloud computing has been used in many applications,and the performance and feasibility of the deployment attracts all attentions of the users,however,the energy optimization is ignored.The application of frequent patterns mining is adopted in this paper.An energy efficient scheculer EEScheduler is developed according to the methods proposed in this paper.And it is deployed to chedule the Map tasks of frequent patterns mining in the cloud to improve the energy efficiency of the system.The experimental results on a cloud platform consisting of 4 nodes show that EEScheduler gains over 60% energy reduction for mining frequent patterns.