An Oracle for Guiding Large-Scale Model/Hybrid Parallel Training of Convolutional Neural Networks

TL;DR

This paper introduces an oracle tool that analyzes and predicts the performance bottlenecks of various parallel training strategies for CNNs, aiding scalable large-scale model training.

Contribution

It provides a model-driven analysis framework and an oracle utility to identify limitations in different parallelism approaches for CNN training at scale.

Findings

Oracle achieves 86.74% accuracy in predicting bottlenecks.

High accuracy of 97.57% for data parallelism.

Evaluated on six strategies, four CNN models, and multiple datasets.

Abstract

Deep Neural Network (DNN) frameworks use distributed training to enable faster time to convergence and alleviate memory capacity limitations when training large models and/or using high dimension inputs. With the steady increase in datasets and model sizes, model/hybrid parallelism is deemed to have an important role in the future of distributed training of DNNs. We analyze the compute, communication, and memory requirements of Convolutional Neural Networks (CNNs) to understand the trade-offs between different parallelism approaches on performance and scalability. We leverage our model-driven analysis to be the basis for an oracle utility which can help in detecting the limitations and bottlenecks of different parallelism approaches at scale. We evaluate the oracle on six parallelization strategies, with four CNN models and multiple datasets (2D and 3D), on up to 1024 GPUs. The results demonstrate that the oracle has an average accuracy of about 86.74% when compared to empirical results, and as high as 97.57% for data parallelism.