Ponte Academic Journal Dec 2016, Volume 72, Issue 12 |
IMPLEMENTATION AND PERFORMANCE ANALYSIS OF KVM-BASED GPU PASS-THROUGH FOR CLOUD DAAS Author(s): Sang Boem Lim ,Guohua Li, Hyeyoung Cho, Jaegyoon Hahm J. Ponte - Dec 2016 - Volume 72 - Issue 12 doi: 10.21506/j.ponte.2016.12.13 Abstract: Desktop as a service (DaaS) is a cloud service that is hosted in the back-end of a virtual desktop infrastructure (VDI). Nowadays, increasing numbers of cloud users are on virtual desktops and most hypervisors support graphics virtualization. However, there are performance issues associated with rendering 3D games and other graphics rendering software on open source hypervisors such as kernel virtual machine (KVM) and Xen. To overcome these problems, we propose using GPU direct pass-through technology to enhance 3D rendering on KVM and to improve the graphics performance of its virtual machines (VMs). In this study, we compared GPU pass-through graphics performance with bare-metal system using two different types of GPU performance benchmark tools and analyzed the performance issues associated with cloud DaaS. The results obtained indicate that direct pass-through does not give the same performance as bare-metal but can be close, virtualization performance is dependent on the number of cores and vertices used, and the appropriate rendering API specific for the job at hand should be used for better performance.
|
Download full text: Check if you have access through your login credentials or your institution |
|
Guide for Authors
This guideline has been prepared for the authors to new submissions and after their manuscripts have been accepted |
Authors Login
We welcome refrees who would be willing to act as reviewers |
Paper Tracking
You can track your submitted article from this tab |
Editorial Board
The international editorial board is headed by Dr. Maria E. Boschi |
General Policies
Papers that are published or held by the Journal may not be published elsewhere |
Peer Review Process
Papers will be sent to three peer reviewers for evaluation |