TensorSocket: Shared Data Loading for Deep Learning Training

Summary: TensorSocket enables collocated DL training processes to share a single data loader, removing redundant CPU preprocessing and data copies while leveraging GPU–GPU interconnects to serve batches directly. Pipeline- and hardware-agnostic, supports heterogeneous models/batch sizes, yields up to 2× throughput and ~50% cloud CPU cost savings, and outperforms CoorDL and Joader. (summarized by gpt-5-mini on Feb 11 2026)

Paper ID: 7532
Venue: SIGMOD
Year: 2026
Pagerank: 5.3483178e-05
Overall Rank: 8,905 | 38.91%
DOI: 10.1145/3749185

Incoming Non-self Citations Over Time

Authors

1. Ties Robroek (IT University of Copenhagen)
2. Neil Kim Nielsen (IT University of Copenhagen)
3. Pınar Tözün (IT University of Copenhagen)

BibTeX Citation

@inproceedings{robroek_sigmod26,
        title = {{TensorSocket: Shared Data Loading for Deep Learning Training}},
        author = {Robroek, Ties and Nielsen, Neil Kim and Tözün, Pınar},
        series = {{SIGMOD} '26},
        booktitle = {Proceedings of the {ACM} {SIGMOD} International Conference on Management of Data},
        publisher = {Association for Computing Machinery},
        doi = {10.1145/3749185},
        url = {https://dl.acm.org/doi/10.1145/3749185},
        year = {2026}
}

Incoming Citations (Sorted by Pagerank)

Showing 1 of 1 citing papers.

Rank	Citing Paper	Year	Venue	Pagerank
10,472	Mixtera: A Data Plane for Foundation Model Training	2026	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.

Rank	Cited Paper	Year	Venue	Pagerank
383	QPipe: A Simultaneously Pipelined Relational Query Engine	2005	SIGMOD	0.00019520728
670	SharedDB: Killing One Thousand Queries With One Stone	2012	VLDB	0.00015157572
1,157	Cerebro: A Data System for Optimized Deep Learning Model Selection	2020	VLDB	0.00011924049
1,446	Analyzing and Mitigating Data Stalls in DNN Training	2021	VLDB	0.0001076818
2,018	tf.data: A Machine Learning Data Processing Framework	2021	VLDB	9.3001937e-05
3,541	Where Is My Training Bottleneck? Hidden Trade-Offs in Deep Learning Preprocessing Pipelines	2022	SIGMOD	7.3280673e-05
3,764	FastFlow: Accelerating Deep Learning Model Training with Smart Offloading of Input Data Pipeline	2023	VLDB	7.1446931e-05
3,782	CROSSBOW: Scaling Deep Learning with Small Batch Sizes on Multi-GPU Servers	2019	VLDB	7.1298979e-05
4,330	Sharing Data and Work Across Concurrent Analytical Queries	2013	VLDB	6.7563122e-05
4,508	To Share or Not to Share?	2007	VLDB	6.6561159e-05
6,024	Expand your Training Limits! Generating Training Data for ML-based Data Management	2021	SIGMOD	6.0031118e-05
8,315	ADDICT: Advanced Instruction Chasing for Transactions	2014	VLDB	5.4553606e-05
8,321	FusionFlow: Accelerating Data Preprocessing for Machine Learning with CPU-GPU Cooperation	2024	VLDB	5.4544352e-05

Semantically Similar Papers

#	Overall Rank	Paper	Year	Venue
1	9,729	Scalable Graph Convolutional Network Training on Distributed-Memory Systems	2023	VLDB
2	1,234	Large Graph Convolutional Network Training with GPU-Oriented Data Communication Architecture	2021	VLDB
3	3,874	Tensors: An abstraction for general data processing	2021	VLDB
4	1,446	Analyzing and Mitigating Data Stalls in DNN Training	2021	VLDB
5	8,321	FusionFlow: Accelerating Data Preprocessing for Machine Learning with CPU-GPU Cooperation	2024	VLDB
6	2,018	tf.data: A Machine Learning Data Processing Framework	2021	VLDB
7	3,541	Where Is My Training Bottleneck? Hidden Trade-Offs in Deep Learning Preprocessing Pipelines	2022	SIGMOD
8	2,473	PyTorch FSDP: Experiences on Scaling Fully Sharded Data Parallel	2023	VLDB
9	521	PyTorch Distributed: Experiences on Accelerating Data Parallel Training	2020	VLDB
10	3,764	FastFlow: Accelerating Deep Learning Model Training with Smart Offloading of Input Data Pipeline	2023	VLDB