Multi-tenant GPU Clusters for Deep Learning Workloads: Analysis and Implications

71 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Pathways: Asynchronous Distributed Dataflow for ML (2022, background): “…optimize for heterogeneity of ML work-loads (Narayanan et al., 2020) and multi-job, multi-user fairness and performance (Xiao et al., 2018; Ren et al., 2015; Mahajan et al., 2020; Jeon et al., 2018), but resources are still exclusively dedicated to single jobs at long time scales (seconds or more).”
Gandiva: Introspective Cluster Scheduling for Deep Learning (2018, methodology): “We run the trace using Hadoop’s YARN capacity scheduler [50] and our Gandiva scheduler.”
Balancing efficiency and fairness in heterogeneous GPU clusters for deep learning (2020, background): “Early schedulers for DLT jobs borrowed directly from the big data schedulers [9, 25].”
Analysis of Large-Scale Multi-Tenant GPU Clusters for DNN Training Workloads (2019, methodology): “We also note that an earlier technical report of our work [26] was used to motivate new scheduling primitives in recent work on scheduling like Gandiva [52].”
Themis: Fair and Efficient GPU Cluster Scheduling for Machine Learning Workloads (2019, background): “However, all ML apps typically compete for GPU resources, and GPU clusters are heavily contended [17].”