{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,6,13]],"date-time":"2026-06-13T04:58:02Z","timestamp":1781326682989,"version":"3.54.1"},"reference-count":63,"publisher":"Association for Computing Machinery (ACM)","issue":"4","funder":[{"DOI":"10.13039\/501100012166","name":"National Key Research and Development Program of China","doi-asserted-by":"publisher","award":["2023YFB4503600"],"award-info":[{"award-number":["2023YFB4503600"]}],"id":[{"id":"10.13039\/501100012166","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/501100001809","name":"National Natural Science Foundation of China","doi-asserted-by":"publisher","award":["U23A20299"],"award-info":[{"award-number":["U23A20299"]}],"id":[{"id":"10.13039\/501100001809","id-type":"DOI","asserted-by":"publisher"}]},{"name":"Beijing Natural Science Foundation","award":["4254083"],"award-info":[{"award-number":["4254083"]}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["Proc. ACM Manag. Data"],"published-print":{"date-parts":[[2025,9,22]]},"abstract":"<jats:p>Learned indexes are viewed as promising substitutes for traditional indexes due to their excellent performance, especially in read-only workloads. Previous studies have shown that updatable learned indexes perform exceptionally well in many cases, suggesting they are nearly ready for real-world applications. However, unlike traditional indexes such as B+tree and ART, updatable learned indexes are prone to instability of real-time trained models, resulting in inherently uncertain structures. This raises skepticism about their robustness, hindering their broader adoption.<\/jats:p>\n                  <jats:p>In this paper, we conduct a systematic benchmark and analysis to address this concern, corroborating doubts about the lack of robustness in state-of-the-art updatable learned indexes. We demonstrate that, contrary to previous findings, updatable learned indexes cannot robustly surpass traditional indexes, even losing their expected advantage under read-intensive workloads. We further reveal the root causes, including overfitted models, unbalanced structures, ineffective adjustments, and excessive space reservation. In addition, we explore potential mitigation methods to address these challenges. We hope our findings will highlight the critical importance of robustness in the design of updatable learned indexes, ultimately paving the way for their real-world adoption.<\/jats:p>","DOI":"10.1145\/3749188","type":"journal-article","created":{"date-parts":[[2025,9,23]],"date-time":"2025-09-23T17:17:03Z","timestamp":1758647823000},"page":"1-25","source":"Crossref","is-referenced-by-count":0,"title":["Understanding Robustness Issues of Updatable Learned Indexes: [Experiments &amp; Analysis]"],"prefix":"10.1145","volume":"3","author":[{"ORCID":"https:\/\/orcid.org\/0009-0003-8361-6766","authenticated-orcid":false,"given":"Yuanhui","family":"Luo","sequence":"first","affiliation":[{"name":"Renmin University of China, Beijing, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-6684-8336","authenticated-orcid":false,"given":"Minhui","family":"Xie","sequence":"additional","affiliation":[{"name":"Renmin University of China, Beijing, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-4573-4676","authenticated-orcid":false,"given":"Yiheng","family":"Tong","sequence":"additional","affiliation":[{"name":"Renmin University of China, Beijing, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0009-0007-5849-7418","authenticated-orcid":false,"given":"Shichao","family":"Jiang","sequence":"additional","affiliation":[{"name":"Renmin University of China, Beijing, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-3286-9259","authenticated-orcid":false,"given":"Yunpeng","family":"Chai","sequence":"additional","affiliation":[{"name":"Renmin University of China, Beijing, China"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"320","published-online":{"date-parts":[[2025,9,23]]},"reference":[{"key":"e_1_2_1_1_1","doi-asserted-by":"publisher","DOI":"10.1145\/3533702.3534917"},{"key":"e_1_2_1_2_1","volume-title":"Rubinstein","author":"Bachfischer Matthias","year":"2022","unstructured":"Matthias Bachfischer, Renata Borovica-Gajic, and Benjamin I. P. Rubinstein. 2022. Testing the Robustness, of Learned Index Structures. arXiv:2207.11575 [cs]"},{"key":"e_1_2_1_3_1","doi-asserted-by":"publisher","DOI":"10.1145\/1734663.1734671"},{"key":"e_1_2_1_4_1","unstructured":"Timo Bingmann. 2013. STX B Tree v0.9. https:\/\/panthema.net\/2007\/stx-btree\/."},{"key":"e_1_2_1_5_1","doi-asserted-by":"publisher","DOI":"10.1145\/3183713.3196896"},{"key":"e_1_2_1_6_1","doi-asserted-by":"publisher","DOI":"10.1145\/3506692"},{"key":"e_1_2_1_7_1","doi-asserted-by":"publisher","DOI":"10.1145\/3654919"},{"key":"e_1_2_1_8_1","doi-asserted-by":"publisher","DOI":"10.1145\/356770.356776"},{"key":"e_1_2_1_9_1","unstructured":"Andrew Crotty. 2021. Hist-Tree: Those Who Ignore It Are Doomed to Learn. In CIDR."},{"key":"e_1_2_1_10_1","first-page":"155","volume-title":"14th USENIX Symposium on Operating Systems Design and Implementation (OSDI 20)","author":"Dai Yifan","year":"2020","unstructured":"Yifan Dai, Yien Xu, Aishwarya Ganesan, Ramnatthan Alagappan, Brian Kroth, Andrea Arpaci-Dusseau, and Remzi Arpaci-Dusseau. 2020. From &#123;WiscKey&#125; to Bourbon: A Learned Index for &#123;Log-Structured&#125; Merge Trees. In 14th USENIX Symposium on Operating Systems Design and Implementation (OSDI 20). 155-171."},{"key":"e_1_2_1_11_1","doi-asserted-by":"publisher","DOI":"10.1145\/3318464.3389711"},{"key":"e_1_2_1_12_1","volume-title":"Proceedings of the 37th International Conference, on Machine Learning. PMLR, 3123-3132","author":"Ferragina Paolo","year":"2020","unstructured":"Paolo Ferragina, Fabrizio Lillo, and Giorgio Vinciguerra. 2020. Why Are Learned Indexes So Effective?. In Proceedings of the 37th International Conference, on Machine Learning. PMLR, 3123-3132."},{"key":"e_1_2_1_13_1","doi-asserted-by":"publisher","DOI":"10.14778\/3389133.3389135"},{"key":"e_1_2_1_14_1","doi-asserted-by":"publisher","DOI":"10.1145\/3299869.3319860"},{"key":"e_1_2_1_15_1","doi-asserted-by":"publisher","DOI":"10.1109\/ICDE55515.2023.00031"},{"key":"e_1_2_1_16_1","doi-asserted-by":"publisher","DOI":"10.1145\/3626752"},{"key":"e_1_2_1_17_1","unstructured":"Daniel Golovin. 2008. Uniquely Represented Data Structures with Applications to Privacy. Ph.D. Dissertation. Carnegie Mellon University."},{"key":"e_1_2_1_18_1","doi-asserted-by":"publisher","unstructured":"Ali Hadian and Thomas Heinis. 2019. Interpolation-Friendly B-trees : Bridging the Gap Between Algorithmic and Learned Indexes. OpenProceedings.org. doi:10.5441\/002\/EDBT.2019.93","DOI":"10.5441\/002\/EDBT.2019.93"},{"key":"e_1_2_1_19_1","volume-title":"MADEX: Learning-augmented Algorithmic Index Structures.. In AIDB@ VLDB.","author":"Hadian Ali","year":"2020","unstructured":"Ali Hadian and Thomas Heinis. 2020. MADEX: Learning-augmented Algorithmic Index Structures.. In AIDB@ VLDB."},{"key":"e_1_2_1_20_1","volume-title":"SOSD: A Benchmark for Learned Indexes. NeurIPS Workshop on Machine Learning for Systems","author":"Kipf Andreas","year":"2019","unstructured":"Andreas Kipf, Ryan Marcus, Alexander van Renen, Mihail Stoian, Alfons Kemper, Tim Kraska, and Thomas Neumann. 2019. SOSD: A Benchmark for Learned Indexes. NeurIPS Workshop on Machine Learning for Systems (2019)."},{"key":"e_1_2_1_21_1","doi-asserted-by":"publisher","DOI":"10.1145\/3401071.3401659"},{"key":"e_1_2_1_22_1","doi-asserted-by":"publisher","DOI":"10.1145\/3514221.3517867"},{"key":"e_1_2_1_23_1","doi-asserted-by":"publisher","DOI":"10.1145\/3183713.3196909"},{"key":"e_1_2_1_24_1","doi-asserted-by":"publisher","DOI":"10.1145\/3588713"},{"key":"e_1_2_1_25_1","volume-title":"IEEE International Conference on Data Engineering (ICDE). IEEE, year=2024 g. 2024. A Fully On-disk Updatable Learned Index. In 40th IEEE International Conference on Data Engineering (ICDE). IEEE.","author":"Lan Hai","unstructured":"Hai Lan, Zhifeng Bao, J Shane Culpepper, Renata Borovica-Gajic, and Yu Don@inproceedingslan2024fully, title=A Fully On-disk Updatable Learned Index, author=Lan, Hai and Bao, Zhifeng and Culpepper, J Shane and Borovica-Gajic, Renata and Dong, Yu, booktitle=40th IEEE International Conference on Data Engineering (ICDE). IEEE, year=2024 g. 2024. A Fully On-disk Updatable Learned Index. In 40th IEEE International Conference on Data Engineering (ICDE). IEEE."},{"key":"e_1_2_1_26_1","doi-asserted-by":"publisher","DOI":"10.14778\/3654621.3654625"},{"key":"e_1_2_1_27_1","doi-asserted-by":"publisher","DOI":"10.1109\/ICDE.2013.6544812"},{"key":"e_1_2_1_28_1","doi-asserted-by":"publisher","DOI":"10.1145\/2933349.2933352"},{"key":"e_1_2_1_29_1","doi-asserted-by":"publisher","DOI":"10.14778\/3489496.3489512"},{"key":"e_1_2_1_30_1","doi-asserted-by":"publisher","DOI":"10.14778\/3598581.3598593"},{"key":"e_1_2_1_31_1","doi-asserted-by":"publisher","DOI":"10.1007\/S00778-024-00893-6"},{"key":"e_1_2_1_32_1","doi-asserted-by":"publisher","DOI":"10.1109\/ICDE.2015.7113282"},{"key":"e_1_2_1_33_1","doi-asserted-by":"publisher","DOI":"10.14778\/3570690.3570704"},{"key":"e_1_2_1_34_1","doi-asserted-by":"publisher","DOI":"10.14778\/3510397.3510405"},{"key":"e_1_2_1_35_1","doi-asserted-by":"publisher","DOI":"10.14778\/3421424.3421425"},{"key":"e_1_2_1_36_1","doi-asserted-by":"publisher","DOI":"10.1145\/3318464.3384706"},{"key":"e_1_2_1_37_1","first-page":"509","volume-title":"Parallel and Distributed Computing and Systems","volume":"509518","author":"McKenney Paul E","year":"1998","unstructured":"Paul E McKenney and John D Slingwine. 1998. Read-copy update: Using execution history to solve concurrency problems. In Parallel and Distributed Computing and Systems, Vol. 509518. Citeseer, 509-518."},{"key":"e_1_2_1_38_1","doi-asserted-by":"publisher","DOI":"10.1145\/3662165.3662764"},{"key":"e_1_2_1_39_1","volume-title":"Technology Conference on Performance Evaluation and Benchmarking. Springer, 17-32","author":"Raman Aneesh","year":"2022","unstructured":"Aneesh Raman, Konstantinos Karatsenidis, Subhadeep Sarkar, Matthaios Olma, and Manos Athanassoulis. 2022. BoDS: A benchmark on data sortedness. In Technology Conference on Performance Evaluation and Benchmarking. Springer, 17-32."},{"key":"e_1_2_1_40_1","doi-asserted-by":"publisher","DOI":"10.1109\/ICDE55515.2023.00117"},{"key":"e_1_2_1_41_1","doi-asserted-by":"publisher","DOI":"10.5555\/3540261.3542552"},{"key":"e_1_2_1_42_1","doi-asserted-by":"publisher","DOI":"10.14778\/3681954.3681969"},{"key":"e_1_2_1_43_1","unstructured":"Florian Scheibner. 2016. ART C. https:\/\/github.com\/flode\/ARTSynchronized."},{"key":"e_1_2_1_44_1","doi-asserted-by":"publisher","DOI":"10.14778\/3594512.3594528"},{"key":"e_1_2_1_45_1","doi-asserted-by":"publisher","DOI":"10.1145\/3332466.3374547"},{"key":"e_1_2_1_46_1","doi-asserted-by":"publisher","DOI":"10.14778\/3461535.3461552"},{"key":"e_1_2_1_47_1","unstructured":"Ziqi Wang. 2017. BTreeOLC. https:\/\/github.com\/wangziqi2016\/index-microbench."},{"key":"e_1_2_1_48_1","doi-asserted-by":"publisher","DOI":"10.14778\/3551793.3551848"},{"key":"e_1_2_1_49_1","doi-asserted-by":"publisher","DOI":"10.14778\/3457390.3457393"},{"key":"e_1_2_1_50_1","doi-asserted-by":"publisher","DOI":"10.1145\/3639293"},{"key":"e_1_2_1_51_1","doi-asserted-by":"publisher","DOI":"10.14778\/3547305.3547322"},{"key":"e_1_2_1_52_1","doi-asserted-by":"publisher","DOI":"10.1145\/3552326.3587434"},{"key":"e_1_2_1_53_1","doi-asserted-by":"publisher","DOI":"10.14778\/3636218.3636232"},{"key":"e_1_2_1_54_1","doi-asserted-by":"publisher","DOI":"10.14778\/3561261.3561270"},{"key":"e_1_2_1_55_1","doi-asserted-by":"publisher","DOI":"10.5555\/3618408.3620112"},{"key":"e_1_2_1_56_1","doi-asserted-by":"publisher","DOI":"10.1145\/2882903.2915222"},{"key":"e_1_2_1_57_1","doi-asserted-by":"publisher","DOI":"10.1145\/3654954"},{"key":"e_1_2_1_58_1","doi-asserted-by":"publisher","DOI":"10.1145\/3639292"},{"key":"e_1_2_1_59_1","doi-asserted-by":"publisher","DOI":"10.14778\/1920841.1920944"},{"key":"e_1_2_1_60_1","doi-asserted-by":"publisher","DOI":"10.1145\/3654948"},{"key":"e_1_2_1_61_1","volume-title":"PLIN: A Persistent Learned Index, for Non-Volatile Memory, with High Performance, and Instant Recovery. VLDB","author":"Zhang Zhou","year":"2023","unstructured":"Zhou Zhang, Zhaole Chu, Peiquan Jin, Yongping Luo, Xike Xie, Shouhong Wan, Yun Luo, Xufei Wu, Peng Zou, Chunyang Zheng, Guoan Wu, and Andy Rudoff. 2023. PLIN: A Persistent Learned Index, for Non-Volatile Memory, with High Performance, and Instant Recovery. VLDB (2023), 13."},{"key":"e_1_2_1_62_1","doi-asserted-by":"publisher","DOI":"10.1145\/3639265"},{"key":"e_1_2_1_63_1","unstructured":"Xinjing Zhou Xiangyao Yu Goetz Graefe and Michael Stonebraker. 2023. Two is better than one: The case for 2-tree for skewed data sets. In CIDR."}],"container-title":["Proceedings of the ACM on Management of Data"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/dl.acm.org\/doi\/pdf\/10.1145\/3749188","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,6,13]],"date-time":"2026-06-13T04:40:42Z","timestamp":1781325642000},"score":1,"resource":{"primary":{"URL":"https:\/\/dl.acm.org\/doi\/10.1145\/3749188"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2025,9,22]]},"references-count":63,"journal-issue":{"issue":"4","published-print":{"date-parts":[[2025,9,22]]}},"alternative-id":["10.1145\/3749188"],"URL":"https:\/\/doi.org\/10.1145\/3749188","relation":{},"ISSN":["2836-6573"],"issn-type":[{"value":"2836-6573","type":"electronic"}],"subject":[],"published":{"date-parts":[[2025,9,22]]}}}