FlashANNS: GPU-Driven Asynchronous I/O Pipelining for Eliminating Storage-Compute Bottlenecks in Billion-Scale Similarity Search
Summary: GPU-driven out-of-core graph ANNS that breaks the SSD/compute bottleneck via dependency-relaxed async pipelining. Query-grained lock-free SSD concurrency plus compute-I/O balanced graph degree selection yield 2.7–12.2x higher throughput at ≥95% recall over DiskANN/SPANN/FusionANNS. (summarized by gpt-5-mini on Apr 11 2026)
Incoming Non-self Citations Over Time
No non-self incoming citations found for this paper in this database.
Authors
- 1. Yang Xiao (Zhejiang University)
- 2. Mo Sun (Zhejiang University)
- 3. Ziyu Song (Zhejiang University)
- 4. Bing Tian (Huazhong University of Science and Technology)
- 5. Jie Sun (Zhejiang University)
- 6. Jie Zhang (Zhejiang University)
- 7. Zeke Wang (Zhejiang University)
- 8. Zonghui Wang (Zhejiang University)
- 9. Wenzhi Chen (Zhejiang University)
- 10. Fei Wu (Zhejiang University)
BibTeX Citation
@inproceedings{xiao_sigmod26,
title = {{FlashANNS: GPU-Driven Asynchronous I/O Pipelining for Eliminating Storage-Compute Bottlenecks in Billion-Scale Similarity Search}},
author = {Xiao, Yang and Sun, Mo and Song, Ziyu and Tian, Bing and Sun, Jie and Zhang, Jie and Wang, Zeke and Wang, Zonghui and Chen, Wenzhi and Wu, Fei},
series = {{SIGMOD} '26},
booktitle = {Proceedings of the {ACM} {SIGMOD} International Conference on Management of Data},
publisher = {Association for Computing Machinery},
doi = {10.1145/3786652},
url = {https://dl.acm.org/doi/10.1145/3786652},
year = {2026}
}Incoming Citations (Sorted by Pagerank)
Showing 0 of 0 citing papers.
| Rank | Citing Paper | Year | Venue | Pagerank |
|---|
Previous
Page 1 / 1
Next
Outgoing Citations (Sorted by Pagerank)
Showing 8 of 8 cited papers.
Citations counted here include only citations to other VLDB/SIGMOD/CIDR/PODS papers in this database.
| Rank | Cited Paper | Year | Venue | Pagerank |
|---|---|---|---|---|
| 21 | Similarity Search in High Dimensions via Hashing | 1999 | VLDB | 0.00056760516 |
| 286 | Milvus: A Purpose-Built Vector Data Management System | 2021 | SIGMOD | 0.00022357911 |
| 1,666 | HippogriffDB: Balancing I/O and GPU Bandwidth in Big Data Analytics | 2016 | VLDB | 0.00010068964 |
| 3,001 | What Modern NVMe Storage Can Do, And How To Exploit It: High-Performance I/O for High-Performance Storage Engines | 2023 | VLDB | 7.8673493e-05 |
| 3,761 | ScaleStore: A Fast and Cost-Efficient Storage Engine using DRAM, NVMe, and RDMA | 2022 | SIGMOD | 7.1450129e-05 |
| 4,763 | Rethinking Logging, Checkpoints, and Recovery for High-Performance Storage Engines | 2020 | SIGMOD | 6.519206e-05 |
| 5,907 | Dotori: A Key-Value SSD Based KV Store | 2023 | VLDB | 6.0453752e-05 |
| 7,261 | LRU-C: Parallelizing Database I/Os for Flash SSDs | 2023 | VLDB | 5.6612613e-05 |
Previous
Page 1 / 1
Next