GPU virtualization support in cloud system

Abstract

Nowadays graphic processing unit (GPU) delivers much better performance than CPU does, and it is becoming increasingly important in high performance computing (HPC) because of its tremendous computing power. At the same time the concept of cloud computing is becoming increasingly popular. This business model suggests that GPU will be more economical because users can spend less money to rent GPUs to fit their special computing needs, rather than buying GPUs. The current practice of virtual GPU rental service is to bind a GPU to a virtual machine statically. As a result this static binding practice is less economical and less flexible. The goal of this paper is to design a GPU provision system that combines CUDA programs from different virtual machines and execute them concurrently, so as to support the concept of GPU sharing among virtual machines. © 2013 Springer-Verlag.