HiPIPE: Adaptive-Chunk Pipeline Scheduling for Hierarchical Collective Communication
Bibliographic record
Abstract
Multi-dimensional networks have become the standard for interconnecting GPUs to support large-scale distributed machine learning (ML) training, driven by increasing computational and memory demands. However, the inherent bandwidth heterogeneity across network dimensions poses a significant challenge in efficiently scheduling collective operations, such as AllReduce, to minimize communication time. Existing hierarchical approaches suffer from inefficiencies due to static or suboptimal pipeline structures with uniform-sized chunks. We propose HiPIPE, a novel solution for multi-dimensional All-Reduce collective communication, incorporating three key optimizations: chunk scheduling across all dimensions, adaptive chunk sizing, and dynamic chunk number selection. These techniques collectively enhance communication efficiency and maximize bandwidth utilization. Particularly, we formulate chunk size optimization as a linear programming problem to derive the optimal chunk sizes, significantly reducing pipeline bubbles and improving overall performance. Extensive experiments using real-testbed emulation demonstrate that HiPIPE consistently outperforms state-of-the-art (SOTA) approaches across diverse network topologies and configurations, achieving an average communication time reduction of 28.85% while sustaining bandwidth utilization of up to 98.11%.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.002 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.003 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".