BatchWeave: A Consistent Object-Store-Native Data Plane for Large Foundation Model Training

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Sun, Ting, Zhang, Junjie, Yan, Xiao, Zhang, Songxin, Song, Zhuoyang, Xi, Jingyi, Mao, Zunyao, Jing, Bingyi, Zhang, Jiaxing, Xie, Zejian
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866909045555724288
author Sun, Ting
Zhang, Junjie
Yan, Xiao
Zhang, Songxin
Song, Zhuoyang
Xi, Jingyi
Mao, Zunyao
Jing, Bingyi
Zhang, Jiaxing
Xie, Zejian
author_facet Sun, Ting
Zhang, Junjie
Yan, Xiao
Zhang, Songxin
Song, Zhuoyang
Xi, Jingyi
Mao, Zunyao
Jing, Bingyi
Zhang, Jiaxing
Xie, Zejian
contents Modern Large Foundation Model (LFM) training has transformed the data pipeline from a static ingestion layer into a dynamic component that must co-evolve with the training process. Existing systems are ill-equipped: colocated dataloaders offer no failure isolation, while message queue-based disaggregated dataloaders operate on a record/offset abstraction that cannot express the batch-level semantics required by distributed training. We present BatchWeave, an object-store-native training data plane for distributed LFM training. BatchWeave uses versioned manifests and conditional object writes to coordinate batch publication, recovery, and lifecycle management. First, it introduces the Transactional Global Batch (TGB), which builds on versioned-manifest ACID storage semantics and extends them with training-specific consistency, including atomic all-rank batch visibility, a globally ordered step sequence, checkpoint-aligned lifecycle management, and end-to-end exactly-once recovery. Second, it realizes recovery and retention directly in the storage layer, by durably persisting producer state through the commit protocol and tying reclamation to distributed checkpoint state. Third, its Decentralized Adaptive Commit (DAC) algorithm sustains stable ingestion throughput as the manifest grows, without any inter-producer communication. Evaluations on large-scale multimodal pre-training and SFT workloads using 64 GPUs show that BatchWeave outperforms colocated dataloader throughput while providing full failure isolation, outperforms Apache Kafka in ingestion throughput, and achieves lower consumer read latency than Kafka.
format Preprint
id arxiv_https___arxiv_org_abs_2605_09994
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle BatchWeave: A Consistent Object-Store-Native Data Plane for Large Foundation Model Training
Sun, Ting
Zhang, Junjie
Yan, Xiao
Zhang, Songxin
Song, Zhuoyang
Xi, Jingyi
Mao, Zunyao
Jing, Bingyi
Zhang, Jiaxing
Xie, Zejian
Distributed, Parallel, and Cluster Computing
Machine Learning
Modern Large Foundation Model (LFM) training has transformed the data pipeline from a static ingestion layer into a dynamic component that must co-evolve with the training process. Existing systems are ill-equipped: colocated dataloaders offer no failure isolation, while message queue-based disaggregated dataloaders operate on a record/offset abstraction that cannot express the batch-level semantics required by distributed training. We present BatchWeave, an object-store-native training data plane for distributed LFM training. BatchWeave uses versioned manifests and conditional object writes to coordinate batch publication, recovery, and lifecycle management. First, it introduces the Transactional Global Batch (TGB), which builds on versioned-manifest ACID storage semantics and extends them with training-specific consistency, including atomic all-rank batch visibility, a globally ordered step sequence, checkpoint-aligned lifecycle management, and end-to-end exactly-once recovery. Second, it realizes recovery and retention directly in the storage layer, by durably persisting producer state through the commit protocol and tying reclamation to distributed checkpoint state. Third, its Decentralized Adaptive Commit (DAC) algorithm sustains stable ingestion throughput as the manifest grows, without any inter-producer communication. Evaluations on large-scale multimodal pre-training and SFT workloads using 64 GPUs show that BatchWeave outperforms colocated dataloader throughput while providing full failure isolation, outperforms Apache Kafka in ingestion throughput, and achieves lower consumer read latency than Kafka.
title BatchWeave: A Consistent Object-Store-Native Data Plane for Large Foundation Model Training
topic Distributed, Parallel, and Cluster Computing
Machine Learning
url https://arxiv.org/abs/2605.09994