TY - GEN
T1 - The Design and Implementation of a High-Performance Hybrid Memory File System
AU - Sha, Edwin H.M.
AU - Chen, Jun
AU - Chen, Xianzhang
AU - Jiang, Weiwen
AU - Zhuge, Qingfeng
N1 - Publisher Copyright:
© 2016 IEEE.
PY - 2017/1/11
Y1 - 2017/1/11
N2 - The emerging Non-Volatile Memory (NVM) connected to the memory bus provides opportunities for high-performance file accesses for its near-DRAM speed and byte-addressability. In this paper, a high-performance hybrid file system, called Hybrid Memory File System (HMFS), is designed and implemented to exploit the high speed of NVM and the large capacity of block devices. A file in HMFS can be stored across the NVM and block devices. In HMFS, each opened file is mapped to a contiguous virtual address space of the kernel. Using the related virtual address space, the physical locations of file data can be efficiently located by the hardware MMU or a dedicated page fault handler for HMFS. The file data in the NVM are accessed with high performance using the corresponding virtual addresses. The file data stored in the block devices are pre-fetched to take advantages of the NVM. Extensive experiments are conducted. The experimental results show that HMFS achieves significant performance improvement over EXT4 with a DRAM cache.
AB - The emerging Non-Volatile Memory (NVM) connected to the memory bus provides opportunities for high-performance file accesses for its near-DRAM speed and byte-addressability. In this paper, a high-performance hybrid file system, called Hybrid Memory File System (HMFS), is designed and implemented to exploit the high speed of NVM and the large capacity of block devices. A file in HMFS can be stored across the NVM and block devices. In HMFS, each opened file is mapped to a contiguous virtual address space of the kernel. Using the related virtual address space, the physical locations of file data can be efficiently located by the hardware MMU or a dedicated page fault handler for HMFS. The file data in the NVM are accessed with high performance using the corresponding virtual addresses. The file data stored in the block devices are pre-fetched to take advantages of the NVM. Extensive experiments are conducted. The experimental results show that HMFS achieves significant performance improvement over EXT4 with a DRAM cache.
UR - https://www.scopus.com/pages/publications/85013177505
U2 - 10.1109/CBD.2016.061
DO - 10.1109/CBD.2016.061
M3 - 会议稿件
AN - SCOPUS:85013177505
T3 - Proceedings - 2016 International Conference on Advanced Cloud and Big Data, CBD 2016
SP - 316
EP - 321
BT - Proceedings - 2016 International Conference on Advanced Cloud and Big Data, CBD 2016
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 4th International Conference on Advanced Cloud and Big Data, CBD 2016
Y2 - 13 August 2016 through 16 August 2016
ER -