high performance computing on graphics processing units: hgpu.org

Similarity search in a large collection of stored objects in a metric database has become a most interesting problem. The Spaghettis is an efficient metric data structure to index metric spaces. However, for real applications processing large volumes of generated data, query response times can be high enough. In these cases, it is necessary to apply mechanisms in order to significantly reduce the average query time. In this sense, the parallelization of metric structures is an interesting field of research. The recent appearance of GPUs for general purpose computing platforms offers powerful parallel processing capabilities. In this paper we propose a GPU-based implementation for Spaghettis metric structure. Firstly, we have adapted Spaghettis structure to GPU-based platform. Afterwards, we have compared both sequential and GPU-based implementation to analyse the performance, showing significant improvements in terms of time reduction, obtaining values of speed-up close to 10.