Analysis of Large-Scale Multi-Tenant GPU Clusters for DNN Training Workloads

TL;DR

This paper analyzes the unique challenges of scheduling and resource utilization in large multi-tenant GPU clusters for deep learning workloads, providing insights and guidelines for improving cluster efficiency.

Contribution

It offers a detailed workload characterization of enterprise GPU clusters and proposes design guidelines for next-generation schedulers tailored to DNN training workloads.

Findings

Gang scheduling and locality constraints impact queuing delays.

Locality significantly affects GPU utilization.

Failures during training influence overall cluster efficiency.

Abstract

With widespread advances in machine learning, a number of large enterprises are beginning to incorporate machine learning models across a number of products. These models are typically trained on shared, multi-tenant GPU clusters. Similar to existing cluster computing workloads, scheduling frameworks aim to provide features like high efficiency, resource isolation, fair sharing across users, etc. However Deep Neural Network (DNN) based workloads, predominantly trained on GPUs, differ in two significant ways from traditional big data analytics workloads. First, from a cluster utilization perspective, GPUs represent a monolithic resource that cannot be shared at a fine granularity across users. Second, from a workload perspective, deep learning frameworks require gang scheduling reducing the flexibility of scheduling and making the jobs themselves inelastic to failures at runtime. In this paper we present a detailed workload characterization of a two-month long trace from a multi-tenant GPU cluster in a large enterprise. By correlating scheduler logs with logs from individual jobs, we study three distinct issues that affect cluster utilization for DNN training workloads on multi-tenant clusters: (1) the effect of gang scheduling and locality constraints on queuing, (2) the effect of locality on GPU utilization, and (3) failures during training. Based on our experience running a large-scale operation, we provide design guidelines pertaining to next-generation cluster schedulers for DNN training workloads.