high performance computing on graphics processing units: hgpu.org

In recent years, graphics processing units (GPUs) have evolved into powerful devices with significant computational performance and memory throughput. Efforts to exploit their potential to tackle problems from various scientific domains with high computational requirements have proven quite successful. In addition, previous research suggests that database query processing algorithms can be accelerated with the utilisation of the GPU. In this project, we use the OpenCL framework to implement a parallel version of a database partitioning algorithm that calculates a set of optimal splitters to partition a database table with specific size bounds. We identify the parts of the algorithm that can be executed in parallel and implement additional steps to combine partial results. By evaluating the performance of the algorithm and how it behaves with input tables of different statistical properties, we conclude that the use of the GPU can result in considerable performance gains, provided certain issues such as memory usage optimisation are addressed.