Development and Evaluation of ParaRuby: a Distributed GPGPU Framework

Abstract

Graphic Processing Unit (GPU), which was entirely used for image processing, has been widely applied to general computation called GPGPU. Even several developing environments are already provided, software developing cost remains high. Implementation of GPGPU program of the target algorithm exploiting parallelism requires not only realization of the target algorithm, but also knowledge of architecture such as memory hierarchy. To provide support for parallel programming with GPGPU, we propose ParaRuby, which is a distributed GPGPU framework using Ruby. The framework enables programmers to implement GPGPU program with Ruby and to execute the program on multiple remote nodes. This paper reports several evaluations of the application implemented on the framework and discusses about performance and programmability.