Recent advances in hardware, such as systems with multiple GPUs and their availability in the cloud, are enabling deep learning in various domains including health care, autonomous vehicles, and Internet of Things. Multi-GPU systems exhibit complex connectivity among GPUs and between GPUs and CPUs. Workload schedulers must consider hardware topology and workload communication requirements in order to allocate CPU and GPU resources for optimal execution time and improved utilization in shared cloud environments. This paper presents a new topology-aware workload placement strategy to schedule deep learning jobs on multi-GPU systems. The placement strategy is evaluated with a prototype on a Power8 machine with Tesla P100 cards, showing speedups of up to ≈1.30x compared to state-of-the-art strategies; the proposed algorithm achieves this result by allocating GPUs that satisfy workload requirements while preventing interference. Additionally, a large-scale simulation shows that the proposed strategy provides higher resource utilization and performance in cloud systems.

This project is supported by the IBM/BSC Technology Center for Supercomputing collaboration agreement. It has also received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No 639595). It is also partially supported by the Ministry of Economy of Spain under contract TIN2015-65316-P and Generalitat de Catalunya under contract 2014SGR1051, by the ICREA Academia program, and by the BSC-CNS Severo Ochoa program (SEV-2015-0493). We thank our IBM Research colleagues Alaa Youssef and Asser Tantawi for the valuable discussions. We also thank SC17 committee member Blair Bethwaite of Monash University for his constructive feedback on the earlier drafts of this paper.

Peer Reviewed