Blink: Fast and Generic Collectives for Distributed ML

Abstract

Model parameter synchronization across GPUs introduces high overheads for data-parallel training at scale. Existing parameter synchronization protocols cannot effectively leverage available network resources in the face of ever increasing hardware heterogeneity. To address this, we propose Blink, a collective communication library that dynamically generates optimal communication primitives by packing spanning trees. We propose techniques to minimize the number of trees generated and extend Blink to leverage heterogeneous communication channels for faster data transfers. Evaluations show that compared to the state-of-the-art (NCCL), Blink can achieve up to 8x faster model synchronization, and reduce end-to-end training time for image classification tasks by up to 40%.

Cited by in corpus (6)

• Communication-Efficient Distributed Deep Learning: A Comprehensive Survey
• Scalable and accurate multi-GPU based image reconstruction of large-scale ptychography data
• Towards Scalable Distributed Training of Deep Learning on Public Cloud Clusters
• Taming Momentum in a Distributed Asynchronous Environment
• Synthesizing Optimal Parallelism Placement and Reduction Strategies on Hierarchical Systems for Deep Learning
• Hoplite: Efficient and Fault-Tolerant Collective Communication for Task-Based Distributed Systems