Minimal vs. non Minimal Adaptive Routing on k-ary n-cubes

There is a common agreement in the scientiic community that ad-aptive routing algorithms will eventually replace the deterministic ones that are currently in use in multicomputer networks. An open issue in this research area is to determine the optimal degree of ad-aptivity that the routing algorithm should provide. Through simulation on a 16-ary 2-cube, we analyze some minimal wormhole algorithms based on Duato's methodology and a non minimal cut-through algorithm, the Chaos routing. The experimental results show that the Chaos routing outperforms the minimal adaptive algorithms under the uniform traac, but suuers from post saturation problems with the complement, transpose and bit reversal traacs. From these results we argue that non minimal algorithms should pose an upper bound on the degree of adaptivity in order to avoid performance degradations due to excessive misrouting.