Increasing the adaptivity of routing algorithms for k-ary n-cubes

Abstract

In this paper, we show that routing algorithms may exploit not only the flexibility obtained by crossing network dimensions in any order but also that obtained in the same network dimension, thanks to the availability of bidirectional channels. We analyze the behavior of adaptive routing algorithms both for deadlock avoidance and recovery, exploiting this increased routing flexibility, and compare them with previous proposals in order to evaluate the contribution of the additional routing freedom on network performance. Simulation results show that this simple improvement in the routing algorithm allows one to achieve throughput improvements of up to 45% in networks with low radix, for a uniform distribution of message destinations.