Evaluation of idealized adaptive routing on k-ary n-cubes

The topology of a k-ary n-cube interconnection network provides multiple paths between a given source and destination pair. Adaptive routing algorithms exploit this feature by allowing messages to choose from among the different paths, in response to changes in system conditions. In this paper, we investigate the performance of an idealized adaptive router on a k-ary n-cube. Using a discrete event network simulator, we characterize and measure the communication speedup of a high bandwidth, fully adaptive algorithm, in order to evaluate the performance gains that can be achieved by any adaptive router. In addition, we develop an analytic queueing model which approximates the behavior of the system. Message latencies are compared to that of e-cube routing, and the impact of various network parameters on system performance is discussed