Comparative analysis of exhaustive searching on a massive finger-vein database over multi-node/multi-core and multi-GPU platforms

When searching on unstructured data (video, images, etc.), response times are a critical factor. In this work we propose an implementation on two types of multi-GPU and multi-node/multi-core platforms, for massive searches. The presented method aims to reduce the time involved in the search process...

Descripción completa

Guardado en:
Detalles Bibliográficos
Autores principales: Guidet, Sebastián, Hernández-García, Ruber, Frati, Fernando Emmanuel, Barrientos, Ricardo J.
Formato: Objeto de conferencia
Lenguaje:Inglés
Publicado: 2022
Materias:
GPU
Acceso en línea:http://sedici.unlp.edu.ar/handle/10915/140636
Aporte de:
Descripción
Sumario:When searching on unstructured data (video, images, etc.), response times are a critical factor. In this work we propose an implementation on two types of multi-GPU and multi-node/multi-core platforms, for massive searches. The presented method aims to reduce the time involved in the search process by solving simultaneous queries over the system and a database of millions of elements. The results show that the multi-GPU approach is 1.6 times superior to the multi-node/multi-core algorithm. Moreover, in both algorithms the speedup is directly proportional to the number of nodes reaching 156x for 4 GPUs, and 87x in the case of the hybrid multi-node/multi-core algorithm.