Review of the Research on the Optimization of the Energy Consumption of the Cloud Platform

Chun-mao Jiang, Yi-bing Li, Li Zhi-cong
2016 International Journal of Smart Home  
Cloud platform is a basic platform to support large data, large-scale and high-frequency access computing. The high-energy consumption of the cloud platform and the schedule of multi-constrained combination in cloud applications is challenging issues faced by the cloud computing. We conducted a systematic review in this paper for the energy consumption of a cloud platform, and pointed out the existing problems, and put forward ideas to solve the above problems by constructing the four-level
more » ... em. Firstly, by conducting virtualization management to the physical resources in the cloud to form the layer of virtual resource. Then, named as presentation layer, to build a formal model of cloud application multi_attribute with effective description, measurement, calculation. Based on this proposed scheduling application layer enables cloud applications to meet the multiple objectives with multi-attribute, heuristic, feedback, iterative scheduling and cloud resources, energy optimization. It laid the foundation for the formation of cloud platform architecture, key technologies and algorithms, which satisfy the multiple objectives: the maximum possible scheduling of the multi-attribute combination constraint application, the optimization for energy consumption and resources in cloud computing. Simply consider the time optimization problem. Optimized time through rational distribution, fuzzy clustering, satisfaction, linear programming, but did not consider the cost of energy consumption, scheduling overhead, reliability, cost, dynamics of tasks, stability, security and so on. Kliazovich, D. et al. [13] focused on the layout method of data between different data centers in the cloud computing environments, reduced time overhead the transmission caused, but it is only discussed from storage resources and network bandwidth, did not save energy though the data layout is more reasonable. YI Kan et al [14] studied and proposed a task distributing algorithm based on load balancing, reduced the response time, but ignored the scheduling overhead, exception handling issues and so on. Lijun Zhang, proposed an earliest completion time scheduling algorithm which used the Multi-service quality value as a priority constraint. For the multi-QoS needs of each user, making unified measurement and then use the metrics of these needs as task precedence constraints, the algorithm reduced the time span in the process of scheduling. Xi Li[15], proposed the concept of deadline satisfaction, the time scheduling algorithm-DSESAW was designed based on satisfaction. According to the priority of subtasks, determined the candidate resources of subtasks during the execution of a workflow. The problem of partitioning the global deadline in the workflow been described as a problem of a constrained nonlinear programming, and solved the problem by the existing methods. Improved the performance of adaptability and time guaranteed aspect, but did not consider the reliability, charges and other service indicators. JianNing Lin① proposed a scheduling algorithm based on genetic algorithm, using the simple coding to get the chromatography relations of the tasks and sorting according to the depth,only considered the execution time of the tasks run on the resources ,and the transmission delay between resources, the dynamic changes of the resource loading and stability are not taken into account. ② proposed a heuristic algorithm based decision path, get the scheduling scheme using the Normal Scheduling Algorithm at first, generating tasks of scheduling decision and the path,then, scheduling the decisive tasks ahead of schedule as soon as possible and running the decisive tasks during the idle periods of resources by using the heuristic algorithm. Achieve the goal to shorten the convergence time of the task, at the same time the same time the method of determining the deadlock been given. Experimental results show that the new algorithm is superior to other heuristic algorithms. Algorithm saves time overhead, but slightly worse than the genetic algorithm in performance. Chen Jing, proposed a scheduling algorithm senior who combined with time constraints, network bandwidth, and the prediction mechanism, the shorter length of the scheduling the higher user satisfaction, achieved a double QOS scheduling, but only to verify the validity of the algorithm does not achieve the specific optimization. Zhi gang chen [11, 16] In order to improve scheduling performance presented a scheduling algorithm multidimensional performance clustering of the resource, constructed the hyper graph of the service's resources at first, clustering around the resources multidimensional, matching the tasks and clustered resources and scheduling, the task completion time shortened, improved performance, but did not consider the aspects of safety and dynamic of the task. Xiao Li Chen[17] proposed a heterogeneous task scheduling algorithm of fuzzy clustering, divided the network resources reasonably and reduced the time of matching resources. With the increasing scale of the task, the high superiority is displayed, but the algorithm does not consider the dynamic changes such as the reliability of links, communication ability. Achieving the time scheduling algorithm basically through the rational layout and the initial linear programming. However, there were lacked of consideration of transmission overhead, dynamic characteristics of the network and multiple performance indicators the in the task scheduling processes. M., et al [18] proposed a fill in the blank copies of the data allocation algorithm. According to the frequency of the agent by which the resource is stored, making a reasonable distribution of the copies of the data. Achieved the optimal efficiency with the minimal overhead. However, the algorithm does not consider computing and throughout capacities of the proxy nodes, and the cost required for transmission, and how to allocate the new data unit. Xin jun Wang, proposed a heuristic algorithm for data distribution, reducing communication costs and improving efficiency, but does not consider the difference of computing power between the nodes. Akon, M., et al. [19, 20] proposed a scheduling algorithm based on P2P phased, the optimal network scheduling scheme, but lack of consideration above service properties. Heidi Liu, proposed a hierarchical genetic algorithm to realize the task scheduling, optimized at the diversity of population, the convergence of the algorithm and the convergence rate, improved the efficiency of genetic algorithm in accuracy and speed, but lack of consideration of the dynamic random characteristics of the tasks. Proposed the task resource allocation map according to the task dependency, according to the optimal selection from the task map, put forward a scheduling algorithm, the algorithm has advantages in case of large data transfers and a large difference of the task resources. Yuan Lin[21], proposed a collaborative filtering algorithm and the concepts of satisfaction, to allocate resources by recommending the resources to a user through the use history, improved the efficiency. However, the algorithm does not consider computing and throughout capacities of the proxy nodes, and cost required for transmission, and without considering how to allocate the new data units, the difference of computing power between nodes, the randomness of the dynamic tasks, the execution of the small amount tasks and so on. Scheduling optimization for resource utilization, the success rate of the service requests, accuracy, etc. The scheduling research for m * n environment, proposed the Nash equilibrium algorithm in the environment, raising the number of the completed tasks in unit time and the average load in networks and systems, but received only an approximate value and does not prove the convergence of the algorithm [14] . Wu Zhiang et al [22] proposed a hierarchical model of QOS, designed a simulation system used to resource management, measuring the QOS parameters on the virtual tissue layer, improved algorithm MIN-MIN, improved the utilization of resources and the success rate of service requests effectively. But the algorithm did not consider other network parameters except the time, the utilization. Dasgupta, P., et al. [23] proposed changed regions activated scheduling algorithm designed for the dynamic scheduling problem in the distributed environments, solved the dynamic scheduling sequence, improved the reuse efficiency and accuracy in designing resources. The schedule about resource utilization did not consider the convergence of the algorithm, time, efficiency, cost, energy consumption and other service expenses. Scheduling algorithm considered the work load and load balancing. Deboosere, L., et al [24] proposed random multi-start climbing algorithm for no center scheduling framework, choice the origin by the exponential growth of the repetitions that were generated in selecting a neighbor randomly, adjusted the workload flexibly, optimized the global node choosing effectively, optimizing the execution performance,
doi:10.14257/ijsh.2016.10.8.23 fatcat:hwbgcqe4sfcntlesjsx5yj2ssu