Saved in:
Bibliographic Details
Main Authors: Luan, Frank Sifei, Wang, Ron Yifeng, Gu, Yile, Mao, Ziming, Lin, Charlotte, Kamsetty, Amog, Chen, Hao, Su, Cheng, Veeramani, Balaji, Lee, Scott, Cho, SangBin, Zinzow, Clark, Liang, Eric, Stoica, Ion, Wang, Stephanie
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2501.12407
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917030919143424
author Luan, Frank Sifei
Wang, Ron Yifeng
Gu, Yile
Mao, Ziming
Lin, Charlotte
Kamsetty, Amog
Chen, Hao
Su, Cheng
Veeramani, Balaji
Lee, Scott
Cho, SangBin
Zinzow, Clark
Liang, Eric
Stoica, Ion
Wang, Stephanie
author_facet Luan, Frank Sifei
Wang, Ron Yifeng
Gu, Yile
Mao, Ziming
Lin, Charlotte
Kamsetty, Amog
Chen, Hao
Su, Cheng
Veeramani, Balaji
Lee, Scott
Cho, SangBin
Zinzow, Clark
Liang, Eric
Stoica, Ion
Wang, Stephanie
contents While ML model training and inference are both GPU-intensive, CPU-based data processing is often the bottleneck. Distributed data processing systems based on the batch or stream processing models assume homogeneous resource requirements. They excel at CPU-based computation but either under-utilize heterogeneous resources or impose high overheads on failure and reconfiguration. We introduce the streaming batch model, a hybrid of batch and streaming that enables efficient and fault-tolerant heterogeneous execution. The key idea is to use partitions as the unit of execution to achieve elasticity, but to allow partitions to be dynamically created and streamed between heterogeneous operators for memory-efficient pipelining. We present Ray Data, a streaming batch system that improves throughput on heterogeneous batch inference pipelines by 2.5-12$\times$ compared to traditional batch and stream processing systems. By leveraging heterogeneous clusters, Ray Data improves training throughput for multimodal models such as Stable Diffusion by 31% compared to single-node ML data loaders.
format Preprint
id arxiv_https___arxiv_org_abs_2501_12407
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle The Streaming Batch Model for Efficient and Fault-Tolerant Heterogeneous Execution
Luan, Frank Sifei
Wang, Ron Yifeng
Gu, Yile
Mao, Ziming
Lin, Charlotte
Kamsetty, Amog
Chen, Hao
Su, Cheng
Veeramani, Balaji
Lee, Scott
Cho, SangBin
Zinzow, Clark
Liang, Eric
Stoica, Ion
Wang, Stephanie
Distributed, Parallel, and Cluster Computing
Machine Learning
While ML model training and inference are both GPU-intensive, CPU-based data processing is often the bottleneck. Distributed data processing systems based on the batch or stream processing models assume homogeneous resource requirements. They excel at CPU-based computation but either under-utilize heterogeneous resources or impose high overheads on failure and reconfiguration. We introduce the streaming batch model, a hybrid of batch and streaming that enables efficient and fault-tolerant heterogeneous execution. The key idea is to use partitions as the unit of execution to achieve elasticity, but to allow partitions to be dynamically created and streamed between heterogeneous operators for memory-efficient pipelining. We present Ray Data, a streaming batch system that improves throughput on heterogeneous batch inference pipelines by 2.5-12$\times$ compared to traditional batch and stream processing systems. By leveraging heterogeneous clusters, Ray Data improves training throughput for multimodal models such as Stable Diffusion by 31% compared to single-node ML data loaders.
title The Streaming Batch Model for Efficient and Fault-Tolerant Heterogeneous Execution
topic Distributed, Parallel, and Cluster Computing
Machine Learning
url https://arxiv.org/abs/2501.12407