材料计算模拟从计算集群到弹性云的迁移研究

发布时间：2018-01-11 09:27

本文关键词：材料计算模拟从计算集群到弹性云的迁移研究　出处：《北京科技大学》2017年博士论文　论文类型：学位论文

【摘要】：理论,计算和实验材料科学与工程的最新进展,不仅是加快新材料发现速度的保证,而且减少了将这些发现作为新产品推向市场所需的时间。利用高通量密度泛函理论(DFT)计算进行新材料的筛选和基础研究,为材料科学和材料创新提供了有趣的设计机会。高通量DFT通常涉及对数万甚至数十万化合物的计算和规模的这种变化需要新的计算能力和数据管理方法。到目前为止,为了确保及时的计算,这种高吞吐量DFT材料模拟必须在专用高性能计算机(HPC)上运行。然而,随着生产和分析材料数据所需的工作数量和复杂程度的增加,HPC环境在合理的时间内解决给定的科学问题成为一个挑战。最新的计算范式,网格和云计算的出现,使科学和IT领域都发生了巨大的变化。一方面,网格计算允许不同种类的资源,以安全的和灵活的方式,从个人电脑到超级计算机进行访问,以解决在科学和工程领域所产生的大规模共享问题。另一方面,云计算是网格的进一步发展,它带来了巨大的机会,以相对较低的成本托管和运行来自不同领域的现实应用,而不需要拥有任何IT基础架构。此外,它还可以通过互联网向消费者提供各种硬件资源作为服务。考虑到在HPC上运行高通量DFT计算的障碍,本文旨在研究在云和网格计算环境中运行高通量DFT材料模拟,以充分利用地理分布和互连的大规模异构来自网格提供的自主资源,同时受益于从动态弹性方式获取大量云资源的可能性。在本论文中,为处理网格和云计算环境中的作业调度问题设计了新算法。以下详细地论述了主要贡献:1)提出了一种有效的作业调度算法,称为两种选择调度算法(TCSA),用于动态分配作业到资源,以最小化作业执行时间,并最大限度地提高云环境中的资源利用率。2)开发了一种改进的粒子群优化算法(PSO)来解决网格环境中的作业调度问题。开发的PSO算法旨在同时最小化最长执行时间的作业调度和所有任务的执行时间3)为解决云计算环境中的作业调度问题设计了基于随机化的负载均衡算法。该算法旨在通过在云中均匀分配资源之间的工作负载,从而最小化作业执行时间并最大限度地提高资源利用率。4)提出了一种简化版本的粒子群优化算法(PSO)来解决云计算环境中的作业调度问题,以完成时间为目标。为了评估所提出的算法的性能,我们通过在不同的场景下进行几个模拟实验,将所提出的作业调度算法与几种现有的最新调度算法相比较。实验结果表明,我们设计的算法工作得很好,能够在合理的时间内找到最优或近似最优解。此外,我们的方法在上述目标方面显著优于其他比较算法,特别是当调度问题变得太复杂或太大时。
[Abstract]:The latest progress of theory calculation and experiment of materials science and engineering, not only speed up new material discovery rate guarantee, but also reduce the findings as new products to the market. The time required to use high-throughput density functional theory (DFT) calculation and selection based on screen for new materials, designed to provide a chance interesting for materials science and materials innovation. The change of high throughput DFT usually involves tens of thousands or even hundreds of thousands of compounds are calculated and the size of the computation ability and data management methods. So far, in order to ensure the timely calculation, the high throughput of DFT material in high performance computer simulation must be special (HPC) on the run. However, with the increase of the number of work required for the production and analysis of material data and the complexity of the HPC environment to solve scientific problems given within a reasonable period of time has become a challenge. New paradigm of computing, grid and cloud computing, has brought about great changes in the field of science and IT. On the one hand, grid computing allows different kinds of resources in a secure and flexible way, from the personal computer to the super computer access, large-scale sharing to solve the issues arising in the fields of Science and engineering. On the other hand, cloud computing is the further development of grid, it has brought great opportunities, practical application of the relatively low cost of hosting and running from different fields, and do not need to have any IT infrastructure. In addition, it can also provide a variety of hardware resources as a service to consumers via the Internet to consider. Operation of high flux DFT in HPC calculation of obstacles, this paper aims to simulate the high flux of DFT material in the cloud and the operation of computing grid environment, in order to make full use of large scale heterogeneous geographic distribution and interconnection From the autonomous resource grid provides the possibility and benefit from access to a large number of cloud resources from the dynamic elastic method. In this paper, a new algorithm is designed for grid and cloud computing scheduling problems in the environment are discussed in detail. The following main contributions: 1) presents a job scheduling algorithm, called two selection scheduling algorithm (TCSA), for the dynamic allocation of jobs to resources, to minimize the execution time of operation, and maximize the utilization of.2 resources in the cloud environment) developed an improved particle swarm optimization (PSO) algorithm to solve the scheduling problem in grid environment. The development of the PSO algorithm at the same time to minimize the execution time of the job scheduling and execution time for all tasks 3) for scheduling problem in cloud computing environment the design of load balancing algorithm based on randomization. It is aimed at. In a uniform distribution of resources between the cloud workloads, thereby minimizing the job execution time and maximize the resource utilization rate of.4) presents a simplified version of the particle swarm optimization (PSO) algorithm to solve the scheduling problem in cloud computing environment, in order to complete the time as the goal. To evaluate the performance of the proposed our algorithm, through several simulation experiments in different scenarios, comparing the proposed scheduling algorithm with several existing new scheduling algorithm. The experimental results show that our algorithm is designed to work well within a reasonable period of time to find the optimal or approximate optimal solution. In addition, our method the target is significantly better than other algorithms, especially when the scheduling problem becomes too complex or too large.

【学位授予单位】：北京科技大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TB305;TP18

【参考文献】