PipeTGL: (Near) Zero Bubble Memory-based Temporal Graph Neural Network Training via Pipeline Optimization

Summary: PipeTGL: pipeline-parallel training for memory-based temporal GNNs that models inter-minibatch memory dependencies via a runtime DAG and uses fine-grained scheduling, operation reordering, and targeted communication to honor chronological memory constraints. Achieves near-zero pipeline bubbles, reduces GPU idle/communication overhead, and delivers 1.27–4.74× speedups with improved multi-GPU training accuracy. (summarized by gpt-5-mini on Feb 09 2026)

Paper ID: 14102
Venue: VLDB
Year: 2025
Pagerank: 5.093636e-05
Overall Rank: 10,887 | 25.31%
DOI: 10.14778/3742728.3742760

Incoming Non-self Citations Over Time

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

Authors

1. Jun Liu (Huazhong University of Science and Technology)
2. Bingqian Du (Huazhong University of Science and Technology)
3. Ziyue Luo (Ohio State University)
4. Sitian Lu (Huazhong University of Science and Technology)
5. Qiankun Zhang (Huazhong University of Science and Technology)
6. Hai Jin (Huazhong University of Science and Technology)

BibTeX Citation

@article{liu_vldb25,
        title = {{PipeTGL: (Near) Zero Bubble Memory-based Temporal Graph Neural Network Training via Pipeline Optimization}},
        author = {Liu, Jun and Du, Bingqian and Luo, Ziyue and Lu, Sitian and Zhang, Qiankun and Jin, Hai},
        journal = {PVLDB},
        series = {{VLDB} '25},
        volume = {18},
        number = {8},
        pages = {2722--2734},
        doi = {10.14778/3742728.3742760},
        url = {https://doi.org/10.14778/3742728.3742760},
        year = {2025}
}

Incoming Citations (Sorted by Pagerank)

Showing 0 of 0 citing papers.

Rank	Citing Paper	Year	Venue	Pagerank

Outgoing Citations (Sorted by Pagerank)

Showing 6 of 6 cited papers.

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

Rank	Cited Paper	Year	Venue	Pagerank
920	APAN: Asynchronous Propagation Attention Network for Real-time Temporal Graph Embedding	2021	SIGMOD	0.00013209734
1,409	TGL: A General Framework for Temporal GNN Training on Billion-Scale Graphs	2022	VLDB	0.00010854808
3,210	Zebra: When Temporal Graph Neural Networks Meet Temporal Personalized PageRank	2023	VLDB	7.6352864e-05
3,634	Orca: Scalable Temporal Graph Neural Network Training with Theoretical Guarantees	2023	SIGMOD	7.2358691e-05
4,878	NeutronStream: A Dynamic GNN Training Framework with Sliding Window for Graph Streams	2024	VLDB	6.4684388e-05
5,205	ETC: Efficient Training of Temporal Graph Neural Networks over Large-scale Dynamic Graphs	2024	VLDB	6.318626e-05

Semantically Similar Papers

#	Overall Rank	Paper	Year	Venue
1	2,640	Scalable and Efficient Full-Graph GNN Training for Large Graphs	2023	SIGMOD
2	3,751	G3: When Graph Neural Networks Meet Parallel Graph Processing Systems on GPUs	2020	VLDB
3	5,597	Accelerating Sampling and Aggregation Operations in GNN Frameworks with GPU Initiated Direct Storage Accesses	2024	VLDB
4	10,330	SWIFT: Enabling Large-Scale Temporal Graph Learning on a Single Machine	2026	SIGMOD
5	7,353	XGNN: Boosting Multi-GPU GNN Training via Global GNN Memory Store	2024	VLDB
6	10,907	Effective and Efficient Distributed Temporal Graph Learning through Hotspot Memory Sharing	2025	VLDB
7	6,730	SIMPLE: Efficient Temporal Graph Neural Network Training at Scale with Dynamic Data Placement	2024	SIGMOD
8	10,521	Efficient GNN Training on Giant Graphs with Collective Batching and Scheduling	2026	VLDB
9	11,110	Towards Ideal Temporal Graph Neural Networks: Evaluations and Conclusions after 10,000 GPU Hours	2025	VLDB
10	1,409	TGL: A General Framework for Temporal GNN Training on Billion-Scale Graphs	2022	VLDB