Distributed synchronized GPU training is commonly used for deep learning. The resource constraint of using a fixed number of GPUs makes large-scale training jobs suffer from long queuing time for resource allocation, and lowers the cluster utilization. Adapting to resource elasticity can alleviate this but often introduces inconsistent model accuracy, due to lacking of capability to decouple model training procedure from resource allocation. We propose EasyScale, an elastic training system that achieves consistent model accuracy under resource elasticity for both homogeneous and heterogeneous GPUs. EasyScale preserves the data-parallel training behaviors strictly, traces the consistency-relevant factors carefully, utilizes the deep learning characteristics for EasyScaleThread abstraction and fast context-switching. To utilize heterogeneous cluster, EasyScale dynamically assigns workers based on the intra-/inter-job schedulers, minimizing load imbalance and maximizing aggregated job throughput. Deployed in an online serving cluster, EasyScale powers the training jobs to utilize idle GPUs opportunistically, improving overall cluster utilization by 62.1%.
@article{arxiv.2208.14228,
title = {EasyScale: Accuracy-consistent Elastic Training for Deep Learning},
author = {Mingzhen Li and Wencong Xiao and Biao Sun and Hanyu Zhao and Hailong Yang and Shiru Ren and Zhongzhi Luan and Xianyan Jia and Yi Liu and Yong Li and Wei Lin and Depei Qian},
journal= {arXiv preprint arXiv:2208.14228},
year = {2023}
}
Comments
To be appeared at SC'23. Link: https://sc23.supercomputing.org/presentation/?id=pap262&sess=sess168