.png){kind=icon}
.png){kind=icon}
Return
WROTree: Write and Range Query Optimal Tree in Persistent Memory
.png){kind=icon}
PRE
AI
DOI：10.1007/978-981-95-5716-5_25.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Persistent memory offers advantages such as byte-addressability, data persistence, and low latency. With Intels introduction of the eADR feature, data persistence has become much easier. However, existing ART variants in persistent memory still suffer from read/write amplification and don't fully exploit eADR. To address these, we propose WROTree (Write and Range Query Optimal Tree): (1) We modify the traditional ART internal node structure by merging partial keys and child node pointers into a combined storage format, thereby reducing write overhead on internal nodes. (2) Leaf nodes use a bidirectional linked list, extract common key prefixes to minimize redundancy, and store the hot fingerprint array in DRAM-all to speed range queries and reduce read/write amplification. (3) It fully exploits eADR by adaptively selecting data persistence methods, thereby enhancing overall performance. Through experimental evaluation, WROTree achieves a 1.2x to 8x improvement in throughput for insertions and range queries compared to state-of-the-art ART-based persistent indexes, and the latency reduced to 0.15x to 0.65x. These results demonstrate WROTree's superior performance in operations such as insertions, updates, and range queries.
Keywords:
Persistent Memory
eADR
ART
Read and Write Amplification
Journal
W
IF:
0
Papers:
35
Citations:
0