Hybrid architecture for metric space searches

Abstract

Every day, new technologies are developed to combine the facilities arranged for shared memory systems with the facilities that provide distributed memory systems. This paper proposes a hybrid system that enables communication between threads running in a shared memory environment and a cluster of computers. To do this we use specific directives provided by MPI to solve a problem of similarity search on metric spaces .This work is part of a larger project that deals with improving query searches over high dimensional spaces, managing large volumes of data, reducing the number of distance evaluations and query response times. While the proposal of this work may be generalized and used for other problems, the results show that the proposed hybrid algorithm allows a significant improvement. This work is part of a larger project that deals with improving the execution of parallel algorithms using a hybrid architecture. The goal is to take advantage of the features and facilities provided by the new parallel architectures that combine distributed and shared memory systems. The former allows to solve large scale problems while the second allows better use of resources.