• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

计算机工程与科学

• 高性能计算 • 上一篇    下一篇

一种支持DDR4的软硬件结合的访存踪迹收集分析工具集

李作骏1,2,陈明宇1,2,秦晓宁3   

  1. (1.中国科学院计算技术研究所,北京 100190;2.中国科学院大学,北京 100049;
    3.曙光信息产业(北京)有限公司,北京 100193)
     
  • 收稿日期:2018-11-16 修回日期:2019-01-04 出版日期:2019-06-25 发布日期:2019-06-25
  • 基金资助:

    国家十三五重点研发计划(2017YFB1001602)

A hybrid memory trace collection
and analysis toolkit for DDR4
 

LI Zuojun1,2,CHEN Mingyu1,2,Qin Xiaoning3
 
  

  1. (1.Institute of Computing Technology,Chinese Academy of Sciences,Beijing 100190;
    2.University of Chinese Academy of Sciences,Beijing 100049;
    3.Dawning Information Industry Co.Ltd.,Beijing 100193,China)
  • Received:2018-11-16 Revised:2019-01-04 Online:2019-06-25 Published:2019-06-25

摘要:

随着多核技术的发展,大数据、云计算、人工智能应用的普及,非易失性内存技术的逐步实用以及信息安全的迫切需求,作为数据处理核心部分的内存系统的设计显得极为重要,而现有的内存系统分析工具却由于各种缺陷已经无法满足研究人员的需求。在原有HMTT的基础上进行硬件级别的重新设计,在最新的DDR41600平台上实现了完整、高效、无失真地获取访存踪迹的功能,并在原有系统的基础上进一步提升了工具的可移植性。最后,使用该工具对最新的SPEC CPU 2017应用进行了访存踪迹的采集测试,并对收集到的访存踪迹信息进行了分析,进一步验证了本文工作的有效性,为今后的各类应用访存行为以及内存系统结构设计研究提供了强有力的工具支撑。
 

关键词: 访存踪迹, FPGA, DDR4, 物理地址

Abstract:

With the development of multicore technology, the popularity of big data, cloud computing, artificial intelligence applications, the gradual practicality of nonvolatile memory technology and the urgent need for information security, the design of memory systems as the core part of data processing is extremely important. However, the existing analysis tools for memory systems are not able to meet the need of the researchers due to a variety of defects. We redesign the hardware based on the original HMTT, and realize the complete, efficient and distortionfree memory trace collection on the latest DDR41600 platform, and further enhance the portability of the toolkit based on the original system. Finally, we use the toolkit to perform memory trace collection test on the latest SPEC CPU 2017 applications, and analyze the collected trace information, which further verifies the effectiveness of our work. Our work offers researchers a powerful tool on analyzing the memory access behaviors of multiple applications and designing memory system architectures.
 

Key words: memory trace, FPGA, DDR4, physical address