Partial migration technique for GPGPU tasks to Prevent GPU Memory Starvation in RPC‐based GPU Virtualization

Abstract

Graphics processing unit (GPU) virtualization technology enables a single GPU to be shared among multiple virtual machines (VMs), thereby allowing multiple VMs to perform GPU operations simultaneously with a single GPU. Because GPUs exhibit lower resource scalability than central processing units (CPUs), memory, and storage, many VMs encounter resource shortages while running GPU operations concurrently, implying that the VM performing the GPU operation must wait to use the GPU. In this paper, we propose a partial migration technique for general-purpose graphics processing unit (GPGPU) tasks to prevent the GPU resource shortage in a remote procedure call-based GPU virtualization environment. The proposed method allows a GPGPU task to be migrated to another physical server's GPU based on the available resources of the target's GPU device, thereby reducing the wait time of the VM to use the GPU. With this approach, we prevent resource shortages and minimize performance degradation for GPGPU operations running on multiple VMs. Our proposed method can prevent GPU memory shortage, improve GPGPU task performance by up to 14%, and improve GPU computational performance by up to 82%. In addition, experiments show that the migration of GPGPU tasks minimizes the impact on other VMs.

Cricket: A virtualization layer for distributed execution of CUDA applications with checkpoint/restart support

“…Kang et al 22 present a technique for migration and partial migration of GPU applications, where a kernel execution can be partially executed on a remote GPU to avoid delays due to memory shortage. They do not create checkpoints that can be restored at a later point in time, but focus on the migration of running applications.…”