Operating Systems Challenges for GPU Resource Management

The graphics processing unit (GPU) is becoming a very powerful platform to accelerate graphics and data-parallel compute-intensive applications. It significantly outperforms traditional multi-core processors in performance and energy efficiency. Its application domains also range widely from embedded systems to high-performance computing systems. However, operating systems support is not adequate, lacking models, designs, and implementation efforts of GPU resource management for multi-tasking environments. This paper identifies a GPU resource management model to provide a basis for operating systems research using GPU technology. In particular, we present design concepts for GPU resource management. A list of operating systems challenges is also provided to highlight future directions of this research domain, including specific ideas of GPU scheduling for realtime systems. Our preliminary evaluation demonstrates that the performance of open-source software is competitive with that of proprietary software, and hence operating systems research can start investigating GPU resource management.