The Case for Learned In-Memory Joins

Summary: First systematic study applying CDF-based learned models and learned indexes (RMI, RadixSpline) to in-memory joins, exploring how model-predicted positions can accelerate INLJ, sort-based, and hash-based joins. Proposes optimized learned variants and shows experimentally that they beat state-of-the-art techniques across many datasets and scenarios. (summarized by gpt-5-mini on Feb 09 2026)

Paper ID: 13034
Venue: VLDB
Year: 2023
Pagerank: 4.5151296e-05
Overall Rank: 8,411 | 41.55%
DOI: 10.14778/3587136.3587148

Incoming Non-self Citations Over Time

Authors

1. Ibrahim Sabek
2. Tim Kraska

Incoming Citations (Sorted by Pagerank)

Showing 5 of 5 citing papers.

Rank	Citing Paper	Year	Venue	Pagerank
8,003	The Holon Approach for Simultaneously Tuning Multiple Components in a Self-Driving Database Management System with Machine Learning via Synthesized Proto-Actions	2024	VLDB	4.6049527e-05
9,322	Are Joins over LSM-trees Ready? Take RocksDB as an Example	2025	VLDB	4.351469e-05
10,217	This is Going to Sound Crazy, But What If We Used Large Language Models to Boost Automatic Database Tuning Algorithms By Leveraging Prior History? We Will Find Better Configurations More Quickly Than Retraining From Scratch!	2026	SIGMOD	4.1905499e-05
10,719	DobLIX: A Dual-Objective Learned Index for Log-Structured Merge Trees	2025	VLDB	4.1905499e-05
10,996	SPID-Join: A Skew-resistant Processing-in-DIMM Join Algorithm Exploiting the Bank- and Rank-level Parallelisms of DIMMs	2024	SIGMOD	4.1905499e-05

Outgoing Citations (Sorted by Pagerank)

Showing 35 of 35 cited papers.

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

Rank	Cited Paper	Year	Venue	Pagerank
52	Database Architecture Optimized for the new Bottleneck: Memory Access	1999	VLDB	0.00066322421
71	How Good Are Query Optimizers, Really?	2016	VLDB	0.00059446482
101	The Case for Learned Index Structures	2018	SIGMOD	0.00049778866
203	Learned Cardinalities: Estimating Correlated Joins with Deep Learning	2019	CIDR	0.00034868567
329	Neo: A Learned Query Optimizer	2019	VLDB	0.00027301488
350	Sort vs. Hash Revisited: Fast Join Implementation on Modern Multi-Core CPUs	2009	VLDB	0.00026368305
403	Multi-Core, Main-Memory Joins: Sort vs. Hash Revisited	2014	VLDB	0.00024176677
538	Design and Evaluation of Main Memory Hash Join Algorithms for Multi-core CPUs	2011	SIGMOD	0.00020632609
584	Massively Parallel Sort-Merge Joins in Main Memory Multi-Core Database Systems	2012	VLDB	0.00019700451
634	Bao: Making Learned Query Optimization Practical	2021	SIGMOD	0.00018844568
729	Umbra: A Disk-Based System with In-Memory Performance	2020	CIDR	0.00017448059
771	Relational Joins on Graphics Processors	2008	SIGMOD	0.00016813054
819	ALEX: An Updatable Adaptive Learned Index	2020	SIGMOD	0.00016237497
844	The PGM-index: a fully-dynamic compressed learned index with provable worst-case bounds	2020	VLDB	0.00015964123
1,077	What happens during a Join? Dissecting CPU and Memory Optimization Effects	2000	VLDB	0.00014224986
1,195	Buffering Accesses to Memory-Resident Index Structures	2003	VLDB	0.00013404419
1,438	Benchmarking Learned Indexes	2021	VLDB	0.00011965956
1,464	Learning Multi-dimensional Indexes	2020	SIGMOD	0.0001184772
1,540	NUMA-aware algorithms: the case of data shuffling	2013	CIDR	0.00011451745
1,608	Qd-tree: Learning Data Layouts for Big Data Analytics	2020	SIGMOD	0.00011169837
1,800	An Experimental Comparison of Thirteen Relational Equi-Joins in Main Memory	2016	SIGMOD	0.00010494121
2,550	Updatable Learned Index with Precise Positions	2021	VLDB	8.5569576e-05
2,777	Quickstep: A Data Platform Based on the Scaling-Up Approach	2018	VLDB	8.1346418e-05
3,179	Asynchronous Memory Access Chaining	2016	VLDB	7.4350144e-05
3,194	Exploiting Coroutines to Attack the "Killer Nanoseconds"	2018	VLDB	7.4068088e-05
3,428	Distributed Join Algorithms on Thousands of Cores	2017	VLDB	7.1002401e-05
3,719	To Partition, or Not to Partition, That is the Join Question in a Real System	2021	SIGMOD	6.8141176e-05
4,056	Are Updatable Learned Indexes Ready?	2022	VLDB	6.4905689e-05
4,086	The Case for a Learned Sorting Algorithm	2020	SIGMOD	6.4579358e-05
5,323	Can Learned Models Replace Hash Functions?	2023	VLDB	5.5671086e-05
5,682	LSched: A Workload-Aware Learned Query Scheduler for Analytical Database Systems	2022	SIGMOD	5.3752251e-05
6,298	Towards instance-optimized data systems	2021	VLDB	5.1182917e-05
6,723	A Critical Analysis of Recursive Model Indexes	2022	VLDB	4.9449538e-05
7,373	Cache-Oblivious Query Processing	2007	CIDR	4.744464e-05
9,784	Is FPGA Useful for Hash Joins? Exploring Hash Joins on Coupled CPU-FPGA Architecture	2020	CIDR	4.2806921e-05

Semantically Similar Papers

Overall Rank	Paper	Year	Venue	Pagerank
2,599	Design and Evaluation of Parallel Pipelined Join Algorithms	1987	SIGMOD	8.4655075e-05
5,323	Can Learned Models Replace Hash Functions?	2023	VLDB	5.5671086e-05
403	Multi-Core, Main-Memory Joins: Sort vs. Hash Revisited	2014	VLDB	0.00024176677
3,664	Processing Sliding Window Multi-Joins in Continuous Queries over Data Streams	2003	VLDB	6.8645092e-05
8,487	Making RDBMSs Efficient on Graph Workloads Through Predefined Joins	2022	VLDB	4.4953604e-05
2,281	Adopting Worst-Case Optimal Joins in Relational Database Systems	2020	VLDB	9.122455e-05
5,723	MCJoin: A Memory-Constrained Join for Column-Store Main-Memory Databases.	2012	SIGMOD	5.3530385e-05
538	Design and Evaluation of Main Memory Hash Join Algorithms for Multi-core CPUs	2011	SIGMOD	0.00020632609
1,800	An Experimental Comparison of Thirteen Relational Equi-Joins in Main Memory	2016	SIGMOD	0.00010494121
1,016	Memory-Efficient Hash Joins	2015	VLDB	0.00014630024