Benchmarking Learned Indexes
Summary: Introduces a unified benchmark for learned indexes, evaluating three learned index families against traditional baselines on four real-world datasets. Finds that learned indexes can outperform non-learned indexes in read-only in-memory dense arrays, and analyzes caching, pipelining, size effects, multi-threading, and build times to explain their performance. (summarized by gpt-5-nano on Feb 09 2026)
Incoming Non-self Citations Over Time
Authors
- 1. Ryan Marcus (Intel; Massachusetts Institute of Technology)
- 2. Andreas Kipf (Massachusetts Institute of Technology)
- 3. Alexander van Renen (Technical University of Munich)
- 4. Mihail Stoian (Technical University of Munich)
- 5. Sanchit Misra (Intel)
- 6. Alfons Kemper (Technical University of Munich)
- 7. Thomas Neumann (Technical University of Munich)
- 8. Tim Kraska (Massachusetts Institute of Technology)
BibTeX Citation
@article{marcus_vldb21,
title = {{Benchmarking Learned Indexes}},
author = {Marcus, Ryan and Kipf, Andreas and van Renen, Alexander and Stoian, Mihail and Misra, Sanchit and Kemper, Alfons and Neumann, Thomas and Kraska, Tim},
journal = {PVLDB},
series = {{VLDB} '21},
volume = {14},
number = {1},
pages = {1--13},
doi = {10.14778/3421424.3421425},
url = {https://doi.org/10.14778/3421424.3421425},
year = {2021}
}Incoming Citations (Sorted by Pagerank)
Showing 9 of 59 citing papers.
| Rank | Citing Paper | Year | Venue | Pagerank |
|---|---|---|---|---|
| 10,458 | From Learning to Recycling: A Log-Structured Learned-Less Index | 2026 | SIGMOD | 5.093636e-05 |
| 10,461 | HIRE: A Hybrid Learned Index for Robust and Efficient Performance under Mixed Workloads | 2026 | SIGMOD | 5.093636e-05 |
| 10,465 | Improving Range Scan Performance in LSM-trees with Group Caching | 2026 | SIGMOD | 5.093636e-05 |
| 10,569 | Toward Drift-Aware Database Benchmarking | 2026 | VLDB | 5.093636e-05 |
| 10,617 | LiBox: A Learned Index as an Array to Minimize Last-Mile Search | 2026 | VLDB | 5.093636e-05 |
| 10,682 | VEGA: An Active-tuning Learned Index with Group-Wise Learning Granularity | 2025 | SIGMOD | 5.093636e-05 |
| 10,955 | DobLIX: A Dual-Objective Learned Index for Log-Structured Merge Trees | 2025 | VLDB | 5.093636e-05 |
| 11,051 | LETIndex: A Secure Learned Index with TEE | 2025 | VLDB | 5.093636e-05 |
| 11,174 | FairHash: A Fair and Memory/Time-efficient Hashmap | 2024 | SIGMOD | 5.093636e-05 |
Outgoing Citations (Sorted by Pagerank)
Showing 13 of 13 cited papers.
Citations counted here include only citations to other VLDB/SIGMOD/CIDR/PODS papers in this database.
Previous
Page 1 / 1
Next
Semantically Similar Papers
| # | Overall Rank | Paper | Year | Venue |
|---|---|---|---|---|
| 1 | 10,277 | On Self-Designing Learned Indexes | 2026 | SIGMOD |
| 2 | 10,378 | High Performance or Low Memory? An Updatable Learned Index Framework for Time-Space Tradeoff | 2026 | SIGMOD |
| 3 | 8,214 | Tuning Hierarchical Learned Indexes on Disk and Beyond | 2022 | SIGMOD |
| 4 | 5,423 | Updatable Learned Indexes Meet Disk-Resident DBMS - From Evaluations to Design Choices | 2023 | SIGMOD |
| 5 | 7,368 | Accelerating String-key Learned Index Structures via Memoization-based Incremental Training | 2024 | VLDB |
| 6 | 9,599 | Understanding Robustness Issues of Updatable Learned Indexes: [Experiments & Analysis] | 2026 | SIGMOD |
| 7 | 43 | The Case for Learned Index Structures | 2018 | SIGMOD |
| 8 | 6,687 | Making In-Memory Learned Indexes Efficient on Disk | 2024 | SIGMOD |
| 9 | 2,806 | Are Updatable Learned Indexes Ready? | 2022 | VLDB |
| 10 | 3,792 | Learned Index: A Comprehensive Experimental Evaluation | 2023 | VLDB |