UNS: A Portable, Mobile, and Exchangeable Namespace for Supporting Fetch-From-Anywhere Big Data Eco-Systems

被引:0
|
作者
Chen, Hsing-bung [1 ]
Guan, Qiang [2 ]
Fu, Song [3 ]
机构
[1] Los Alamos Natl Lab, UltraScale Syst Res Ctr, Los Alamos, NM 87544 USA
[2] Kent State Univ, Dept Comp Sci, Kent, OH 44240 USA
[3] Univ North Texas, Dept Comp Sci & Engn, Denton, TX 76203 USA
关键词
Big Data Computing; Data Sharing; File system; Object storage; Namespace; Metadata; Universal Data Access;
D O I
10.1109/DASC/PiCom/DataCom/CyberSciTec.2018.00-12
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Modern global file systems such as parallel file systems or distributed file systems normally use a global namespace (GNS) to support viewing and accessing files that are independent of their physical storage locations. Using GNS support, we can seamlessly modify and reconfigure physical data storage without affecting how users view and access it. However, with the rapid growth of data size, the current design and implementations of the GNS system is experiencing problems of manageability and scalability. In this paper, we present a new namespace architecture called the Universal Namespace (UNS). UNS enhances both the local and global namespace with portable, parallel data mobile, and exchangeable properties and supports fetch-from-anywhere storage systems without requiring remote mounting. UNS exploits the extended file attributes of file systems and enhances file system metadata with unified remote data access information and methods. We design the proposed UNS to support big data computing on multiple geolocation-based shared storage systems. We describe the innovative concepts of the proposed UNS's architecture and demonstrate UNS's early performance results from various test cases. Finally, we summarize the benefits and advantages of using UNS in supporting remote data sharing for extreme scale computing and big data computing. Furthermore, we present the future design and developmental direction of UNS.
引用
收藏
页码:889 / 896
页数:8
相关论文
empty
未找到相关数据