Multi-resource interleaving for deep learning training

Training Deep Learning (DL) model requires multiple resource types, including CPUs, GPUs, storage IO, and network IO. Advancements in DL have produced a wide spectrum of models that have diverse usage patterns on different resource types. Existing DL schedulers focus on only GPU allocation, while missing the opportunity of packing jobs along multiple resource types. We present Muri, a multi-resource cluster scheduler for DL workloads. Muri exploits multi-resource interleaving of DL training jobs to achieve high resource utilization and reduce job completion time (JCT). DL jobs have a unique staged, iterative computation pattern. In contrast to multi-resource schedulers for big data workloads that pack jobs in the space dimension, Muri leverages this unique pattern to interleave jobs on the same set of resources in the time dimension. Muri adapts Blossom algorithm to find the perfect grouping plan for single-GPU jobs with two resource types, and generalizes the algorithm to handle multi-GPU jobs with more than two types. We build a prototype of Muri and integrate it with PyTorch. Experiments on a cluster with 64 GPUs demonstrate that Muri improves the average JCT by up to 3.6× and the makespan by up to 1.6× over existing DL schedulers.