Procesamiento de búsquedas por similitud. Tecnologías de Paralelización e Indexación

The similarity search consists on retrieving all objects within a database that are similar or relevant to a particular query. Nowadays, it has become a topic of great interest to the scientific community because of its multiple fields of application, such as searching for words and images on the World Wide Web, pattern recognition, plagiarism detection, multimedia databases, among others. Search by similarity or proximity is modeled mathematically through a metric space in which its objects are represented as black boxes where the only information available is the distance of one object to the rest. In general, calculating the function of distance is costly, and search systems operate at a high rate of queries per time unit. To optimize this process, numerous metric structures have been developed; such structures work as indexes of the data and preprocess it in order to reduce the distance evaluations during the search. Furthermore, the need to process large volumes of data makes it impractical to use a structure in real application environments if it does not consider the use of parallel processing environments. There are several technologies for parallel processing implementations. Technologies based on multi-CPU (multi-core) and GPU / multi-GPU architectures are amongst the most current and interesting due to high performance and low costs involved. The next Scientific-Technical Report addresses the similarity search and the implementation of metric structures on parallel environments. It also presents the state of the art related to similarity search on metric structures and parallelism technologies. Comparative analysis are also proposed, seeking to identify the behavior of a set of metric spaces and metric structures over processing platforms multicore-based and GPU-based.