Development and evaluation of processor farm algorithms/software for robot sensing applications

Using a parallel multi-transputer pipeline hardware architecture, this paper analyzes the processor farm paradigm for robot sensing applications. Algorithms for parallel search in a database, parallel sorting and parallel image processing were developed. We show how the processor farm is a dynamic load balancing software depending on factors which are automatically determined by the system. Considering software modifications in the farmer, like the activity of the sender or the use of buffers, we have evaluated the performance of several variants of farm algorithms. The results, obtained in real situations, give us indications for the best choice of the farmer software and the optimum number of worker processors