摘要
本文提出了一种多核处理器自适应I/O直接缓存访问(ADCA)的方法以提升I/O访存的性能,降低对其他程序的影响。与传统直接缓存访问(DCA)不同的是,该方法利用了LRU栈特性,通过采样辅助标签目录的方式动态调整DCA可使用的cache空间,同时对I/O数据的替换和写内存策略进行优化。实验结果表明,与DCA方式相比,该方式使得I/O带宽提升了大约10%,而与SPEC和采用直接内存访问(DMA)方式的网络测试程序同时运行相比,SPEC定点和浮点性能分别提升了11. 5%和8. 9%。
This paper presents a method of adaptive direct cache access（ ADCA） for chip multi-core processors to improve memory access performance of I/O device and reduce the impact on other programs. Unlike traditional direct cache access（ DCA）,this approach takes advantage of the LRU stack property to dynamically adjust the cache space available to DCA by sampling the auxiliary tag directory,and simultaneously optimizes the replacement strategy and write memory strategy for I/O data. The experimental results show that compared with the DCA method,the proposed method improves the I/O bandwidth by about 10%. Compared with SPEC running with network in the direct memory access（ DMA） method,SPECint rate and SPECfp rate gain by 11. 5% and 8. 9% respectively.
作者
李鹏
曾露
王焕东
Li Peng;Zeng Lu;Wang Huandong(State Key Laboratory of Computer Architecture（Institute of Computing Technology,Chinese Academy of Sciences）,Beijing 100190;Institute of Computing Technology,Chinese Academy of Sciences,Beijing 10019;University of Chinese Academy of Sciences,Beijing 10004;Loongson Technology Corporation Limited,Beijing 10019)
出处
《高技术通讯》
EI
CAS
北大核心
2018年第6期488-499,共12页
Chinese High Technology Letters
基金
国家"核高基"科技重大专项课题(2009ZX01028-002-003
2009ZX01029-001-003
2014ZX01020201
2014ZX01030101)
国家自然科学基金(61521092
61232009
61222204
61432016)
中国科学院重点部署(ZDRW-XH-2017-1)资助项目
