Distributed routing in computer communication networks

This paper considers a class of simple distributed shortest-path routing algorithms first introduced in the ARPA Network with convergence for fixed cost functions proven by W.D. Tajibnapis. The basic algorithm has since been implemented in somewhat modified form by a number of operating networks as well as in some computer communication architectures. The basic algorithm, although simple to implement, is prone to loops and utilizes O(n3) control packets during its convergence mode. A so-called Predecessor algorithm is described that retains simplicity of implementation yet reduces the incidence of loops and the number of control packets generated. Comparative theoretical studies of the two algorithms are presented for full-duplex loop networks. Comparative performance results of the two algorithms for large networks, using computer experimentation, are presented as well.