GPU Virtualization
Bring Your Own Device (BYOD) will be popular in the business world, so mobile virtualization is also hot issue to isolate business/private environments. On the other hand, performance gap between GPU and CPU in the mobile environment will grow, so guest will have several limitations without GPU from perspective of performance.
It is a new GPU virtualization solution on KVM/ARM platform based on device file boundary, and guest/KVM host communicate each other by hypervall (hvc). In this project, we support jb-mr1 android graphics systems such as ion, fb, vsync.
Our GPU resource sharing model is foreground/background. In this case, we can stop layer composition of background surfaceflinger. User can use input trigger to change foreground/background state.
Demo
publications
- 박병수, 이운성, 박세진, 박찬익, 김우성, 조혜진, 이규은.“안드로이드 호스트 기반 KVM/ARM 가상화 환경에서의 GPU 디바이스 구동기법” 한국 컴퓨터 종합 학술 대회 (Korea Computer Congress) 2014