high performance computing on graphics processing units: hgpu.org

Data analysis is a rising field of interest for computer science research due to the growing amount of information that is digitally available. This increase in data has as direct consequence that any analysis is significantly complex. By using structured representations for the data sets, like graphs, the analysis becomes feasible, but is still time-consuming. In this project, the focus is on the reduction of the computational time for data analysis, with the introduction of accelerators. Accelerators are specialized hardware components that assist the general processing unit in performing (parts of) the task at hand. In particular, we focus on the use of General Purpose Graphical Processing Units (GPGPU) to help speedup the analysis. GPUs are specifically designed for representing and manipulating graphical data which invoke the processing of large chunks of data, GPUs are designed with large numbers of concurrent processing units and thus have a high potential of improving performance. In this project, we show the impact of using GPGPUs for both simple and more complex analysis, varying from small to large data sets, with the use of a programming model called OpenCL. We compare the performance of using accelerators against the traditional CPU-based implementation. Due to the inter-platform portability of the OpenCL model, such comparison can be performed without having to alter the algorithm. The use of accelerators is expected to become beneficial for analysis that require large computational power. For example, search algorithms (that require little to no computation) are not expected to profit from accelerators, while the more complex, centrality analysis is expected to have significantly more benefit from accelerators. Our results clearly shows this shift of performance improvement when algorithms further utilize the potential of accelerators, because the analysis grows in size and/ or complexity.