Authors: Linchen Yu; Hong Yao; Xiaofei Liao
Addresses: School of Computer Science and Technology, China University of Geosciences, Hubei Wuhan, 430074, China ' School of Computer Science and Technology, China University of Geosciences, Hubei Wuhan, 430074, China ' School of Computer Science and Technology, Huazhong University of Science and Technology, Hubei Wuhan, 430074, China
Abstract: Virtual machine (VM) technologies offer lots of benefits such as users' isolation, server consolidation and live migration. However, owing to the overhead incurred by indirect access to physical resources such as GPU, IO devices and VM technologies have not been widely used in high performance computing area. General purpose graphics processing unit (GPGPU) computing solution for virtual machines makes it possible that high performance computing (HPC) applications running on GPU can be ported to virtual machines (VMs) environment. A novel GPU resources management system built on VMs called VMGPURMS and a novel GPU cluster schedule policy are presented to use VM as computing node, and enable users to run jobs in the whole virtual machine environment transparently. Evaluation demonstrates that the efficiency of GPUs with the GPU cluster schedule policy could be improved by 17% over the traditional cluster schedule policy without taking GPU into consideration.
Keywords: GPGPU; general purpose GPU; graphics processing unit; CUDA virtualisation; virtual machines; CPU/GPU hybrid systems; GPU scheduling; GPU resources; resource management; high performance computing; cluster scheduling.
International Journal of High Performance Computing and Networking, 2016 Vol.9 No.5/6, pp.423 - 430
Available online: 22 Nov 2016 *Full-text access for editors Access for subscribers Purchase this article Comment on this article