Improving the user experience of the rCUDA remote GPU virtualization framework

摘要

Graphics processing units (GPUs) are being increasingly embraced by the high-performance computingrncommunity as an effective way to reduce execution time by accelerating parts of their applications. remoternCUDA (rCUDA) was recently introduced as a software solution to address the high acquisition costs andrnenergy consumption of GPUs that constrain further adoption of this technology. Specifically, rCUDA is arnmiddleware that allows a reduced number of GPUs to be transparently shared among the nodes in a cluster.rnAlthough the initial prototype versions of rCUDA demonstrated its functionality, they also revealed concernsrnwith respect to usability, performance, and support for new CUDA features. In response, in this paper,rnwe present a new rCUDA version that (1) improves usability by including a new component that allowsrnan automatic transformation of any CUDA source code so that it conforms to the needs of the rCUDArnframework, (2) consistently features low overhead when using remote GPUs thanks to an improved newrncommunication architecture, and (3) supports multithreaded applications and CUDA libraries. As a result,rnfor any CUDA-compatible program, rCUDA now allows the use of remote GPUs within a cluster with lowrnoverhead, so that a single application running in one node can use all GPUs available across the cluster,rnthereby extending the single-node capability of CUDA.