Taming Unbalanced Training Workloads in Deep Learning with Partial Collective Operations
This addresses inefficiencies in distributed training systems for deep learning practitioners, offering a practical improvement over existing methods.
The paper tackles load imbalance in distributed deep learning training by proposing eager-SGD, which relaxes global synchronization using partial collective operations, achieving a 1.27x speedup over synchronous SGD without accuracy loss on datasets like CIFAR-10 and ImageNet.
Load imbalance pervasively exists in distributed deep learning training systems, either caused by the inherent imbalance in learned tasks or by the system itself. Traditional synchronous Stochastic Gradient Descent (SGD) achieves good accuracy for a wide variety of tasks, but relies on global synchronization to accumulate the gradients at every training step. In this paper, we propose eager-SGD, which relaxes the global synchronization for decentralized accumulation. To implement eager-SGD, we propose to use two partial collectives: solo and majority. With solo allreduce, the faster processes contribute their gradients eagerly without waiting for the slower processes, whereas with majority allreduce, at least half of the participants must contribute gradients before continuing, all without using a central parameter server. We theoretically prove the convergence of the algorithms and describe the partial collectives in detail. Experimental results on load-imbalanced environments (CIFAR-10, ImageNet, and UCF101 datasets) show that eager-SGD achieves 1.27x speedup over the state-of-the-art synchronous SGD, without losing accuracy.