ParvaGPU: Efficient Spatial GPU Sharing for Large-Scale DNN Inference in Cloud Environments

TL;DR

This paper proposes ParvaGPU, a technology that facilitates spatial GPU sharing for large-scale DNN inference in cloud computing and addresses the challenges of minimizing underutilization within allocated GPU space partitions and external fragmentation in combined MIG and MPS environments.

Abstract

In cloud environments, GPU-based deep neural network (DNN) inference servers are required to meet the Service Level Objective (SLO) latency for each workload under a specified request rate, while also minimizing GPU resource consumption. However, previous studies have not fully achieved this objective. In this paper, we propose ParvaGPU, a technology that facilitates spatial GPU sharing for large-scale DNN inference in cloud computing. ParvaGPU integrates NVIDIA's Multi-Instance GPU (MIG) and Multi-Process Service (MPS) technologies to enhance GPU utilization, with the goal of meeting the diverse SLOs of each workload and reducing overall GPU usage. Specifically, ParvaGPU addresses the challenges of minimizing underutilization within allocated GPU space partitions and external fragmentation in combined MIG and MPS environments. We conducted our assessment on multiple A100 GPUs, evaluating 11 diverse DNN workloads with varying SLOs. Our evaluation revealed no SLO violations and a significant reduction in GPU usage compared to state-of-the-art frameworks.

Figures (10)

• Figure 1: Supported MIG configurations on the NVIDIA A100 GPU.
• Figure 2: Overall design of ParvaGPU.
• Figure 3: Throughput (requests/s) of InceptionV3 with different batch sizes and instance sizes for each process count of 1 (a), 2 (b), and 3 (c).
• Figure 5: Total number of GPUs of each baseline and ParvaGPU.
• Figure 6: Internal slack rate of each baseline and ParvaGPU.