Technologies of snapshots. (a) Snapshots based on COW friendly B-tree. (b) Snapshots of HMVFS. (c) Snapshots of NOVA-Fortis.

Download： Full Size Img PowerPoint

Technologies of snapshots. (a) Snapshots based on COW friendly B-tree. (b) Snapshots of HMVFS. (c) Snapshots of NOVA-Fortis.

2023, 38(2): 348-372.

A Survey of Non-Volatile Main Memory File Systems .

Figures of the Article

Architecture of several major NVMM file systems. (a) Traditional I/O architecture. (b) NVMM FS architecture. (c) Bypassing VFS architecture. (d) User-level architecture. (e) Using cache architecture.
File data block index structure of (a) ext3, (b) SCMFS, and (c) SIMFS.
File system bandwidth when preforming random (a) read and (b) write operations.
Latency of metadata operations. stat, create, rename and delete use system calls and perform reading, creating, renaming and deleting a file respectively. (a) Cold cache. (b) Warm cache.
NUMA architecture with two NUMA nodes. APP represents application.
Metadata scalability on ext4-dax. ops/s means the number of operations per second.
Metadata scalability of creating file.
Impact of consistency scalability. N and D represent running on NVMM and DRAM respectively.
Consistency technology of COW and journaling. Inode block (InA) stores file inodes; indirect index block (Ind) stores directory index; block stores file data; new data: new written data in file system; additional data: extra data caused by writing new data. (a) Traditional COW technique. (b) NVMM COW technique. (c) Traditional journaling technique. (d) NVMM journaling technique.
Consistency technique of log-structuring.
Technology of soft updates. (a) Traditional soft updates. (b) Soft updates of SoupFS.
Technologies of snapshots. (a) Snapshots based on COW friendly B-tree. (b) Snapshots of HMVFS. (c) Snapshots of NOVA-Fortis.
Ratio of random read/write latency to sequential read/write latency respectively.
Normalized latency of Optane PMM and Optane SSD against DRAM. 256 B and 16 KB represent the access block size. A and S represent Optane PMM and SSD respectively. (a) Read latency. (b) Write latency.

Others

Chinese PDF
2023-2-10-1054-Chinese Information 27KB
English PDF
2023-2-10-1054-Highlights 146KB
Read Online
https://rdcu.be/dhR6a

Related articles

Impacts of Dirty Data on Classification and Clustering Models: An Experimental Evaluation

2021, 36(4): 806-821. DOI: 10.1007/s11390-021-1344-6
Subgroup Discovery Algorithms: A Survey and Empirical Evaluation

2016, 31(3): 561-576. DOI: 10.1007/s11390-016-1647-1
Performance Evaluation of Machine Learning Methods in Cultural Modeling

2009, 24(6): 1010-1017.
Server-Based Data Push Architecture for Multi-Processor Environments

2007, 22(5): 641-652.
An Action Analysis for Combining Partial Evaluation

2000, 15(2): 196-201.
Demand Priority Protocol Simulation and Evaluation

1999, 14(6): 599-605.
NONH:A New Cache-Based Coherence Protocol for Linked List Structure DSM System and Its Performance Evaluation

1996, 11(4): 405-415.
A Form Evaluation System and Its Data Structure for Brush-Written Chinese Characters

1995, 10(1): 35-41.
Bottom-up Evaluation of Datalog with Negation

1994, 9(3): 229-244.
On Parallel Evaluation of Ordered Attribute Grammars

1991, 6(4): 347-354.