摘要
为提升物联网与边缘计算应用中前端节点间的数据访问效率,提出了一种新型远程零拷贝文件系统。该文件系统无须借助特殊硬件,可直接基于通用网卡设备实现零拷贝的数据传输框架;充分利用新型非易失内存的随机访问特性,尽可能减少数据缓存和拷贝,提高数据访问的吞吐量。建立缓冲区池,精简并融合传统网络栈和存储栈,从而缩短文件访问路径,减少软件冗余,降低数据访问延迟。最终提供高带宽、低延迟的远程数据访问性能。测试结果表明,远程零拷贝文件系统比传统网络文件系统减少了42.26%~99.19%的读写延迟,细粒度访问下的吞吐量最高可提升1297倍,显著降低了处理器开销和缓存失效次数。
In order to enhance data access efficiency between front-end nodes of Internet of things and edge computing applications,a novel RZCFS(remote zero-copy file system)was proposed.The zero-copy data transmission framework based on generic network interface card was realized without special hardware.The random-access character of non-volatile memory was fully utilized by RZCFS to reduce data caches and copies,which yields a significant throughput improvement for data access.A buffer pool was implemented,the traditional network stack and storage stack were simplified and converged to shorten the file access path,which can eliminate software overhead and reduce data access latency.As a result,RZCFS provides the low-latency and high-bandwidth remote data access.Simulation results show that it achieves 42.26%~99.19%latency reductions for traditional NFS(network file system).The throughput of fine-grained access can reach 1297 times faster than those of NFS.The RZCFS significantly reduces the processor cycles and cache misses.
作者
韩文炳
陈小刚
李顺芬
李大刚
陈诗雁
段有康
宋志棠
HAN Wenbing;CHEN Xiaogang;LI Shunfen;LI Dagang;CHEN Shiyan;DUAN Youkang;SONG Zhitang(State Key Laboratory of Functional Materials for Informatics, Shanghai Institute of Microsystem and Information Technology, Chinese Academy of Sciences, Shanghai 200050, China;University of Chinese Academy of Sciences, Beijing 100049, China;School of Electronic and Computer Engineering, Peking University Shenzhen Graduate School, Shenzhen 518055, China)
出处
《国防科技大学学报》
EI
CAS
CSCD
北大核心
2020年第3期9-16,共8页
Journal of National University of Defense Technology
基金
国家重点研发计划资助项目(2017YFA0206101,2017YFB0701703,2017YFA0206104,2018YFB0407500,SQ2017YFGX020134)
国家自然科学基金资助项目(61874129,61874178,61504157,61622408)
中国科学院战略性先导科技专项资助项目(XDPB12)
上海市科委资助项目(17DZ2291300,18DZ2272800)。
关键词
新型非易失内存
物联网
边缘计算
零拷贝
网络文件系统
non-volatile memory
Internet of things
edge computing
zero copy
network file system