文章摘要
李作骏,卢天越,朱晨光,陈明宇.一种软硬件协同的访存踪迹实时分析处理工具[J].高技术通讯(中文),2025,35(8):813~823
一种软硬件协同的访存踪迹实时分析处理工具
A hardware and software collaborative tool for real-time memory trace analysis and processing
  
DOI:10. 3772 / j. issn. 1002-0470. 2025. 08. 001
中文关键词: 内存访问; 踪迹; 实时分析; 软硬件协同; 安全监测
英文关键词: memory access, trace, real-time analysis, software-hardware collaboration, security monitoring
基金项目:
作者单位
李作骏 (中国科学院计算技术研究所处理器芯片全国重点实验室北京 100190) (中国科学院大学计算机科学与技术学院北京 100190) 
卢天越  
朱晨光  
陈明宇  
摘要点击次数: 88
全文下载次数: 66
中文摘要:
      本文提出一种软硬件协同的访存踪迹实时分析处理工具,通过硬件采集并实时回传访存踪迹、实时分析处理访存踪迹、提取语义信息即时指导访存系统调控,解决了当前基于访存踪迹信息进行访存调控的软硬件系统研究中缺乏有效的实时分析处理手段的问题。该工具通过定制现场可编程门阵列(field programmable gate array,FPGA)采集卡,在双路Intel E5 2680 v4服务器平台中实现同机访存踪迹回传和实时分析处理。基准测试表明该工具的最低传输延迟为1.5μs,能够实时处理DDR4-1600速率下的高带宽访存,在安全监测场景实验中的平均异常检测响应时间为28.0μs,基于该工具的实时远端内存预取方案性能在25.0%和50.0%本地内存负载下分别比Fastswap高24.9%和32.0 %。
英文摘要:
      In this paper, a real-time analysis and processing tool is proposed for memory traces with hardware and software collaboration, which solves the problem of lacking effective real-time analysis and processing methods for the research of memory access management mechanisms based on real-time memory trace. The tool provides hardware memory trace collection and real-time feedback, on-site memory trace analysis and semantic information extraction, and instant guidance for memory system tuning. A prototype system is implemented in a dual-socket Intel E5 2680 v4 server platform with customized field programmable gate array (FPGA) collection card. Benchmark tests show that the prototype has a minimum transfer latency of 1.5μs, and can handle full-bandwidth memory accesses at DDR4-1600 rate in real time. It achieves an average anomaly detection response time of 28.0μs in security monitoring scenario experiments, and the performance of a real-time remote memory prefetching scheme based on the tool is 24.9% and 32.0% higher than Fastswap under 25.0% and 50.0% local memory load, respectively.
查看全文   查看/发表评论  下载PDF阅读器
关闭

分享按钮