基于微处理器的并行计算系统的优化及应用研究

发布时间：2018-05-04 21:36

本文选题：微处理器 + 并行计算系统的优化　；参考：《吉林师范大学》2013年硕士论文

【摘要】：随着材料科学技术的发展与第一原理的应用，模拟计算需要处理的数据量与日俱增，人类对计算能力的需求与芯片的发展速度已经明显的不匹配，仅通过提高单个处理器的运算速度和采用传统的串行计算技术已难以解决上述问题。为此基于微处理器计算系统及并行计算技术应运而生，同时也为我们提供了一种高速计算的解决方案。由于大型的并行系统集群的价格十分昂贵，中小型的科研机构无力承担，因此利用普通微机处理器搭建并行计算平台成为解决这一难题的有效手段。基于微处理器的并行计算系统因其廉价、灵活易维护的特点逐渐应用到模拟计算等多个领域。本文主要利用微处理器这一优势和特点，通过反复的实验研究，成功的搭建了基于微处理器的并行计算系统，并通过实例的运行，对比了并行计算和串行计算的运行时间，凸显了并行计算优势及高效性。在系统搭建的过程中，针对系统存在的问题，进行了以下三方面优化，，提出了三种无密码通信的方案，为提高通信安全及通信效率方面提供了借鉴。为降低多结点并行计算系统的部署及维护成本，采用PXE网络启动技术构建出了基于微处理器的无盘并行计算系统，使其更易于管理和维护。本文还实现了单机多核并行计算系统的构建，充分发挥多核微处理器的性能，可以提高并行系统的各节点间的通信效率，有效地解决了多机并行系统通信效率低的瓶颈问题。本文的另一主要工作是在搭建好的并行计算平台上实现了Siesta并行编译，并将其应用于分子的结构模拟计算中，从而验证了基于微处理器的并行系统的可用性及其高效性。综上所述，采用微处理器构建的并行计算系统具有高性能机群的相应特点和优势，可以被广泛的应用于中小企业及科研机构中，为其提供了理想计算平台。
[Abstract]:With the development of material science and technology and the application of first principle, the amount of data that needs to be processed in simulation calculation is increasing day by day. The demand for computing power and the development speed of chip have been obviously mismatched. It is difficult to solve these problems only by increasing the speed of a single processor and adopting the traditional serial computing technology. Therefore, microprocessor based computing system and parallel computing technology emerge as the times require, and also provide us with a high-speed computing solution. Because the price of large-scale parallel system cluster is very expensive and the small and medium-sized scientific research institutions can not afford it, it is an effective means to solve this problem by using the common microcomputer processor to build a parallel computing platform. The parallel computing system based on microprocessor has been applied to many fields such as analog computation because of its low cost, flexibility and maintainability. Based on the advantages and characteristics of microprocessor, this paper successfully builds a parallel computing system based on microprocessor through repeated experimental research, and compares the running time of parallel computing and serial computing through an example. The advantages and efficiency of parallel computing are highlighted. In the course of the system construction, the following three aspects of optimization are carried out in view of the problems existing in the system, and three schemes of cipher free communication are put forward, which can be used for reference in improving communication security and communication efficiency. In order to reduce the cost of deployment and maintenance of multi-node parallel computing system, a diskless parallel computing system based on microprocessor is constructed by using PXE network startup technology, which makes it easier to manage and maintain. This paper also realizes the construction of a single-machine multi-core parallel computing system, which can give full play to the performance of the multi-core microprocessor. It can improve the communication efficiency between nodes of the parallel system, and effectively solve the bottleneck problem of the low communication efficiency of the multi-computer parallel system. Another main work of this paper is to implement Siesta parallel compilation on a good parallel computing platform, and apply it to molecular structural simulation, which verifies the availability and high efficiency of parallel system based on microprocessor. To sum up, the parallel computing system based on microprocessor has the corresponding characteristics and advantages of high performance cluster, which can be widely used in small and medium-sized enterprises and scientific research institutions, and provides an ideal computing platform for it.
【学位授予单位】：吉林师范大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP338.6

【相似文献】