移动云计算环境下多DAG工作流的节能调度算法及能耗测量研究

发布时间：2018-04-24 20:45

本文选题：移动云计算 + 数据中心　；参考：《新疆大学》2014年硕士论文

【摘要】：随着近年来移动互联网的快速发展，云计算的概念被引入移动网络，从而构造了新型云服务模型—移动云计算。海量接入终端、爆炸性增长的计算任务和移动终端电量受限等特点给移动云计算的发展带来了新挑战。本文着重关注移动云环境下的全局节能问题。首先，近年来数据中心能耗的急剧增高使全球用电压力骤然上升，大量研究致力于提高数据中心能效比，降低空载数据中心服务器的比例；其次，由于移动设备电量的局限性，移动端设备节能也成为近年来的研究热点。大量研究表明，可将移动设备的计算任务上传到资源服务器（例如，附近其它计算节点或云端数据中心）以减小终端用户的计算能耗。然而，在某些特定场景下，云端数据中心的服务可用性较差。因此，具有较强灵活性的移动私有云被广泛研究。产生计算任务的移动设备可利用其网络拓扑附近的多台其它移动设备完成存在依赖关系的并行任务工作流集群，以达到节能计算的目标。有向无环图（DAG）能反映出工作流中各任务间的相互约束关系，因而被普遍应用于讨论面向并行任务集群的工作流调度问题。然而，随着移动应用和集群任务数量的爆炸式增长，现有基于单DAG工作流的调度方案有如下问题：（1）在多种形态的多DAG工作流复杂场景下无法兼顾所有情况和全面考虑整体节能效果；（2）对于多DAG工作流的性能优化效果尚可，却带来了大量的能耗，无法兼顾性能提升与能耗优化；（3）适用场景局限性较大，算法普适性较差。针对上述问题，，本文分别针对移动云环境下的数据中心节能和移动私有云环境下的终端设备节能提出了两种基于多DAG工作流的节能调度算法—MREO和EAMRS。MREO以减少数据中心处理器的占用数目为目的，整合独立任务后，进一步对计算密集型和通信密集型的任务特点进行分析，并采用回溯和分支限界进行整合路径的动态优化选择。EAMRS算法合理利用工作流中各任务间的松弛时间，减少了多DAG工作流下移动计算节点的占用数目，通过任务复制有效控制传输能耗，以降低移动私有云的整体能耗。此外，文章采用了兼顾使用率、尾功耗和网络影响因子的智能手机能耗评估模型，提高了算法评估的准确性和有效性。模拟实验表明，两种算法分别在移动云数据中心和移动私有云场景下有效降低了系统的计算和传输能耗，确保了多DAG工作流的性能，并采用PowerMonitor等能耗测量设备较准确的评估了算法的能耗优化有效性。
[Abstract]:With the rapid development of mobile Internet in recent years, the concept of cloud computing has been introduced into mobile network, and a new cloud service model, mobile cloud computing, has been constructed. The characteristics of massive access terminals, explosively growing computing tasks and limited power of mobile terminals have brought new challenges to the development of mobile cloud computing. This paper focuses on global energy conservation in mobile cloud environment. First, the sharp increase in energy consumption in data centers in recent years has led to a sharp rise in global power pressure. A great deal of research has been devoted to improving the energy efficiency ratio of data centers and reducing the proportion of no-load data center servers; secondly, due to the limitations of mobile devices, In recent years, energy saving of mobile devices has become a hot topic. A large number of studies have shown that computing tasks of mobile devices can be uploaded to resource servers (for example, other computing nodes in the vicinity or cloud data centers) to reduce the computing energy consumption of end users. However, in some specific scenarios, the service availability of cloud data centers is poor. Therefore, mobile private cloud with strong flexibility has been widely studied. Mobile devices that generate computing tasks can use several other mobile devices near their network topology to complete a dependent parallel task workflow cluster in order to achieve the goal of energy-saving computing. The directed acyclic graph (DAG) can reflect the inter-constraint relation among the tasks in the workflow, so it is widely used to discuss the workflow scheduling problem for the parallel task cluster. However, as the number of mobile applications and cluster tasks explodes, The existing scheduling scheme based on single DAG workflow has the following problem: 1) under the complex scenario of multi-form multi- workflow, it can not take into account all situations and consider the overall energy-saving effect.) the performance optimization effect of multi- workflow can be achieved. However, it brings a lot of energy consumption, which can not take into account the performance improvement and energy consumption optimization. In response to the above problems, In this paper, we propose two energy-saving scheduling algorithms, MREO and EAMRS.MREO, for data center energy saving in mobile cloud environment and terminal device energy saving in mobile private cloud environment, respectively, in order to reduce the number of data center processors. After integrating independent tasks, the characteristics of computation-intensive and communication-intensive tasks are further analyzed, and the dynamic optimization selection of integration path using backtracking and branch limits. EAMRS algorithm makes reasonable use of the relaxation time between tasks in workflow. In order to reduce the total energy consumption of mobile private cloud, the number of mobile computing nodes in multi- workflow is reduced, and the transmission energy consumption is effectively controlled by task replication. In addition, a smart phone energy consumption evaluation model which takes account of utilization rate, tail power consumption and network impact factors is used to improve the accuracy and effectiveness of the algorithm evaluation. Simulation experiments show that the two algorithms can effectively reduce the calculation and transmission energy consumption of the system in the mobile cloud data center and the mobile private cloud scene, and ensure the performance of the multi- workflow. The energy consumption optimization efficiency of the algorithm is evaluated accurately by using PowerMonitor and other energy consumption measurement equipment.
【学位授予单位】：新疆大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.09

【参考文献】