基于多核技术的并行计算平台实现与应用

发布时间：2018-11-27 19:11

【摘要】：大众多核化的今天,多核处理器在人们的日常生活中随处可见。多核处理器就是在一个处理器上集成两个或多个运算核心,在并行性能提升的情况下,相对于单核它能够以更低的频率处理更高的工作负载,这样就能够在低功耗、小散热量以及低成本的情况下提升处理器的性能。处理器从单核发展到多核,传统的、适用于单核的串行编程模式与多核的并行架构之间的矛盾越发明显。为了充分利用多核性能,就需要设计一种并行编程模型来编写更高效的应用程序。另一方面,多媒体和网络技术的迅速发展与普及。这种数据量大、对实时性要求高、计算处理具有高度可并行的任务,在传统串行架构上优化的有限性,给多核的并行化带来了新的机遇。如何在多核环境下更好的发挥处理器的性能优势,满足图像和流媒体应用的需求就成了当前面临的主要困难。为了解决这个问题,本文设计和实现了一种新的基于多核架构的并行计算平台,应用于大计算量的程序性能的优化。它以任务为最小的运行单位,便于程序员将现有的程序改成并行的程序。同时使用任取窃取结合动态负载均衡算法,使任务更加均衡的分配到多核处理器的各个计算核心上,更加有效的利用处理器的计算资源,从而使并行性和计算性能得到很大的提升。本文的主要贡献如下：(1)在详细剖析了多核处理器体系结构和并行算法特点的基础上,分析影响多核并行计算性能的主要因素,探讨适合于发挥多核性能的算法和编程技术。(2)提出了一种基于任务的、同时结合负载平衡算法的任务窃取多核并行计算平台模型。(3)实现了一个针对多媒体和图像等大数据计算的并行计算平台。
[Abstract]:Nowadays, multicore processors are widely used in people's daily life. A multicore processor integrates two or more computing cores on a single processor, which can handle higher workloads at a lower frequency than a single core when parallel performance is improved, thus enabling low power consumption. Small heat loss and low cost to improve processor performance. The processor has developed from single core to multi core, and the contradiction between the traditional serial programming mode and the parallel architecture of multi core is becoming more and more obvious. In order to make full use of multi-core performance, we need to design a parallel programming model to write more efficient applications. On the other hand, the rapid development and popularization of multimedia and network technology. Because of the large amount of data, the high requirement of real-time, the high parallelism of computing processing, the finiteness of optimization in the traditional serial architecture brings new opportunities for the parallelization of multi-cores. How to give full play to the performance advantage of the processor in multi-core environment and meet the needs of image and streaming media applications has become the main difficulty. In order to solve this problem, this paper designs and implements a new parallel computing platform based on multi-core architecture, which is applied to the optimization of program performance of large computational complexity. It is the task of the smallest running unit, easy for programmers to change existing programs into parallel programs. At the same time, the use of arbitrary theft combined with dynamic load balancing algorithm, so that tasks more evenly distributed to the various computing cores of multi-core processors, more effective use of processor computing resources, Therefore, parallelism and computing performance are greatly improved. The main contributions of this paper are as follows: (1) based on the detailed analysis of the architecture of multi-core processors and the characteristics of parallel algorithms, the main factors affecting the performance of multi-core parallel computing are analyzed. This paper discusses algorithms and programming techniques suitable for multi-core performance. (2) A task-based algorithm is proposed. At the same time, a multi-core parallel computing platform model based on load balancing algorithm is proposed. (3) A parallel computing platform for big data computing such as multimedia and image is implemented.
【学位授予单位】：天津科技大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP338.6

【相似文献】