Abstract

An approach to mapping structure of parallel program onto structure of distributed computer system by the Hopfield neural network is presented. For typical structures of parallel programs ("line", "ring", "mesh", "binary tree") and regular structures of distributed CS ("torus", "hypercube") it is shown that fuzzy control of derivative of neuron activation function can essentially improve the mapping performance.

File
tarkov.pdf157.58 KB
Issue
Pages
83-88