Study of a parallel CBIR implementation using MPI

This paper presents a parallel implementation of a content based information retrieval (CBIR) system which deals with an image database composed of data from over 29 million bidimensional RGB images, which would be equivalent to 1.45 TB of graphical data. The application has been designed for a distributed memory multiprocessor environment, and has been implemented in a cluster of twenty five PCs using MPI. The paradigm that best fits the problem's needs is a farm based solution: a master process distributes the work load between the slave processes, and when these have finished, the master recollects the partial results computed on each slave process. In order to evaluate this solution, the experimental results have been compared with those achieved using a Silicon Graphics Origin 2000, a shared memory machine with eight processors. This paper analyzes the performances offered by both approaches from the viewpoints of speed, price and scalability, presenting the conclusions that can be extracted from the results' comparison.