Intra-node Memory Safe GPU Co-Scheduling

TL;DR

This paper introduces schedGPU, a framework for safe intra-node GPU co-scheduling that significantly improves GPU utilization and application performance by sharing GPUs among multiple applications while respecting memory constraints.

Contribution

The paper presents schedGPU, a novel framework with four scheduling policies that enables safe GPU sharing among multiple applications considering memory constraints, validated on real-world workloads.

Findings

Over 10x GPU utilization improvement for single applications

Up to 5x speed-up in total execution time for multi-application workloads

Increased average GPU utilization and memory utilization by 5 and 12 times

Abstract

GPUs in High-Performance Computing systems remain under-utilised due to the unavailability of schedulers that can safely schedule multiple applications to share the same GPU. The research reported in this paper is motivated to improve the utilisation of GPUs by proposing a framework, we refer to as schedGPU, to facilitate intra-node GPU co-scheduling such that a GPU can be safely shared among multiple applications by taking memory constraints into account. Two approaches, namely a client-server and a shared memory approach are explored. However, the shared memory approach is more suitable due to lower overheads when compared to the former approach. Four policies are proposed in schedGPU to handle applications that are waiting to access the GPU, two of which account for priorities. The feasibility of schedGPU is validated on three real-world applications. The key observation is that a performance gain is achieved. For single applications, a gain of over 10 times, as measured by GPU utilisation and GPU memory utilisation, is obtained. For workloads comprising multiple applications, a speed-up of up to 5x in the total execution time is noted. Moreover, the average GPU utilisation and average GPU memory utilisation is increased by 5 and 12 times, respectively.