Rethinking Learned Index and LSM-tree Integration

Summary: Identifies mismatches between learned-index assumptions and LSM compaction: SSTable sizing matters less, while fixed error bounds fail under shifting distributions. Wild Turkey co-tunes level-aware compaction and RL-controlled error bounds, improving throughput and reducing stalls and compactions. (summarized by gpt-5.6-luna on Aug 28 2026)

Paper ID: h20f4089d934595af
Venue: VLDB
Year: 2026
Pagerank: 4.9769913e-05
Overall Rank: 10,854 | 27.05%
DOI: 10.14778/3836663.3836671
PDF: Download (CC BY-NC-ND 4.0)

Incoming Non-self Citations Over Time

No non-self incoming citations found for this paper in this database.

Authors

1. Guangxun Zhao (Dankook University)
2. Yongjie Zhu (Dankook University)
3. Charles Jaranilla (Dankook University)
4. Seehwan Yoo (Dankook University)
5. Jongmoo Choi (Dankook University)

BibTeX Citation

@article{zhao_vldb26,
        title = {{Rethinking Learned Index and LSM-tree Integration}},
        author = {Zhao, Guangxun and Zhu, Yongjie and Jaranilla, Charles and Yoo, Seehwan and Choi, Jongmoo},
        journal = {PVLDB},
        series = {{VLDB} '26},
        volume = {19},
        number = {11},
        pages = {3034--3047},
        doi = {10.14778/3836663.3836671},
        url = {https://doi.org/10.14778/3836663.3836671},
        year = {2026}
}

Incoming Citations (Sorted by Pagerank)

Showing 0 of 0 citing papers.

Rank	Citing Paper	Year	Venue	Pagerank

Outgoing Citations (Sorted by Pagerank)

Showing 19 of 19 cited papers.

Citations counted here include only citations to other VLDB/SIGMOD/CIDR/PODS papers in this database.

Rank	Cited Paper	Year	Venue	Pagerank
40	The Case for Learned Index Structures	2018	SIGMOD	0.00046363107
258	bLSM: A General Purpose Log Structured Merge Tree	2012	SIGMOD	0.00022932099
400	Monkey: Optimal Navigable Key-Value Store	2017	SIGMOD	0.00019124757
422	ALEX: An Updatable Adaptive Learned Index	2020	SIGMOD	0.00018488849
436	Optimizing Space Amplification in RocksDB	2017	CIDR	0.00018312911
458	The PGM-index: a fully-dynamic compressed learned index with provable worst-case bounds	2020	VLDB	0.00017880664
555	SageDB: A Learned Database System	2019	CIDR	0.0001650754
753	Dostoevsky: Better Space-Time Trade-Offs for LSM-Tree Based Key-Value Stores via Adaptive Removal of Superfluous Merging	2018	SIGMOD	0.00014232329
768	FITing-Tree: A Data-aware Index Structure	2019	SIGMOD	0.00014107655
835	Benchmarking Learned Indexes	2021	VLDB	0.00013575971
1,525	Updatable Learned Index with Precise Positions	2021	VLDB	0.00010355133
2,582	Are Updatable Learned Indexes Ready?	2022	VLDB	8.2641447e-05
2,861	APEX: A High-Performance Learned Index on Persistent Memory	2022	VLDB	7.9295281e-05
2,868	Spooky: Granulating LSM-Tree Compactions Correctly	2022	VLDB	7.9205462e-05
3,233	TreeLine: An Update-In-Place Key-Value Store for Modern Storage	2023	VLDB	7.5008192e-05
3,621	Learned Index: A Comprehensive Experimental Evaluation	2023	VLDB	7.1510804e-05
6,123	Learning to Optimize LSM-trees: Towards A Reinforcement Learning based Key-Value Store for Dynamic Workloads	2023	SIGMOD	5.8771312e-05
8,540	AirIndex: Versatile Index Tuning Through Data and Storage	2023	SIGMOD	5.3192281e-05
10,007	BT-Tree: A Reinforcement Learning Based Index for Big Trajectory Data	2024	SIGMOD	5.0954911e-05

Semantically Similar Papers

#	Overall Rank	Paper	Year	Venue
1	5,378	Updatable Learned Indexes Meet Disk-Resident DBMS - From Evaluations to Design Choices	2023	SIGMOD
2	10,500	On Self-Designing Learned Indexes	2026	SIGMOD
3	835	Benchmarking Learned Indexes	2021	VLDB
4	6,123	Learning to Optimize LSM-trees: Towards A Reinforcement Learning based Key-Value Store for Dynamic Workloads	2023	SIGMOD
5	6,934	Breaking Down Memory Walls in LSM-based Storage Systems	2020	SIGMOD
6	8,382	Tuning Hierarchical Learned Indexes on Disk and Beyond	2022	SIGMOD
7	3,621	Learned Index: A Comprehensive Experimental Evaluation	2023	VLDB
8	6,613	Making In-Memory Learned Indexes Efficient on Disk	2024	SIGMOD
9	9,140	Rethinking The Compaction Policies in LSM-trees	2025	SIGMOD
10	10,656	From Learning to Recycling: A Log-Structured Learned-Less Index	2026	SIGMOD