high performance computing on graphics processing units: hgpu.org

The most popular heterogeneous many-core platform, the CPU+GPU combination, has received relatively little attention in operating systems research. This platform is already widely deployed: GPUs can be found, in some form, in most desktop and laptop PCs. Used for more than just graphics processing, modern GPUs have proved themselves versatile enough to be adapted to other applications as well. Though GPUs have strengths that can be exploited in systems software, this remains a largely untapped resource. We argue that augmenting the OS kernel with GPU computing power opens the door to a number of new opportunities. GPUs can be used to speed up some kernel functions, make other scale better, and make it feasible to bring some computation-heavy functionality into the kernel. We present our framework for using the GPU as a co-processor from an OS kernel, and demonstrate a prototype in Linux.