非对称多核体系下的阿姆达尔定律性能模型研究

发布时间：2018-08-05 13:45

【摘要】：阿姆达尔定律已经成为了并行计算性能模型中最为重要的定理之一,它简明清晰地论述了多核体系所带来的性能提升,以及在并行环境下应用程序的性能瓶颈所在,影响了后续一系列计算机理论的发展。并行理论一般基于对称多核体系进行研究,即所有的核都拥有同样的运算能力。但根据阿姆达尔定律的推导,此体系的并行任务性能最终会受限于串行部分所占的比例。为了寻求突破这瓶颈的方法,处理器领域中出现了非对称多核体系的新研究方向。但从对称架构转到非对称架构时,应该如何设计使得非对称架构的效果最佳,却缺少一定的理论分析、指导和实验。本文基于非对称多核体系架构,提出了一种将任务的串并行比例、底层核的具体配置等因素加入阿姆达尔定律的建模方法,通过量化两架构各自核的性能,分析架构之间的性能差异,进一步肯定了非对称体系的任务加速比优势。该理论模型同时发现当对称架构转化为非对称架构之时,应符合一定的设计转换原则才能最优化整个系统。模型在综合考虑加速比性能、硬件平台转换的开销以及上层任务串并行比例后,给出了架构转换的指导思想。本文的实验平台是multi2sim多核模拟器,通过修改源代码后使其支持了非对称多核的体系架构,并在该实验环境下运行相关的基准程序,验证了理论模型的部分结论,为后续的非对称架构发展和研究打下基础。
[Abstract]:Amdal's law has become one of the most important theorems in parallel computing performance models. It gives a concise and clear description of the performance improvements brought about by multicore systems and the performance bottlenecks of applications in parallel environments. It affects the development of a series of computer theories. Parallel theory is generally based on symmetric multicore systems, that is, all cores have the same computing power. But according to Amdal's law, the parallel task performance of this system will be limited by the proportion of serial part. In order to find a way to break through this bottleneck, a new research direction of asymmetric multicore system appears in processor field. However, when moving from symmetric architecture to asymmetric architecture, how to design the asymmetric architecture to make the optimal effect, but the lack of theoretical analysis, guidance and experiment. Based on the asymmetric multicore architecture, this paper proposes a modeling method that adds the factors such as the parallel ratio of tasks and the specific configuration of the underlying core to the Amdal law, and quantifies the performance of the respective cores of the two architectures. By analyzing the performance differences between architectures, the task speedup advantage of asymmetric systems is further affirmed. At the same time, it is found that when the symmetric architecture is transformed into an asymmetric architecture, the system can be optimized according to a certain design transformation principle. After considering the speedup performance, the overhead of the hardware platform transformation and the parallel ratio of the upper task string, the model gives the guiding idea of the architecture transformation. The experimental platform of this paper is multi2sim multi-core simulator. By modifying the source code, it supports the architecture of asymmetric multicore, and runs the related benchmark program in this experimental environment, which verifies some conclusions of the theoretical model. Lay the foundation for the further development and research of asymmetric architecture.
【学位授予单位】：上海交通大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP338.6

【相似文献】