BAGUA: Scaling up Distributed Learning with System Relaxations

TL;DR

BAGUA is a flexible MPI-style library that enables advanced system relaxations for distributed training, significantly improving training speed and adaptability across diverse network conditions.

Contribution

It introduces a modular MPI-style communication library supporting various system relaxation techniques for distributed learning.

Findings

BAGUA outperforms PyTorch-DDP, Horovod, and BytePS by up to 2x in training time.

It demonstrates effective use of system relaxations like quantization and decentralization.

Tradeoff analysis shows different algorithms excel under different network conditions.

Abstract

Recent years have witnessed a growing list of systems for distributed data-parallel training. Existing systems largely fit into two paradigms, i.e., parameter server and MPI-style collective operations. On the algorithmic side, researchers have proposed a wide range of techniques to lower the communication via system relaxations: quantization, decentralization, and communication delay. However, most, if not all, existing systems only rely on standard synchronous and asynchronous stochastic gradient (SG) based optimization, therefore, cannot take advantage of all possible optimizations that the machine learning community has been developing recently. Given this emerging gap between the current landscapes of systems and theory, we build BAGUA, a MPI-style communication library, providing a collection of primitives, that is both flexible and modular to support state-of-the-art system relaxation techniques of distributed training. Powered by this design, BAGUA has a great ability to implement and extend various state-of-the-art distributed learning algorithms. In a production cluster with up to 16 machines (128 GPUs), BAGUA can outperform PyTorch-DDP, Horovod and BytePS in the end-to-end training time by a significant margin (up to 2 times) across a diverse range of tasks. Moreover, we conduct a rigorous tradeoff exploration showing that different algorithms and system relaxations achieve the best performance over different network conditions.