Globally Adaptive Load-Balanced Routing on k-ary n-cubes

We introduce a new method of adaptive routing on k-ary ncubes that we refer to as Globally Adaptive Load-balance (GAL). Unlike previous adaptive routing algorithms that make routing decisions based on local information (typically channel queue depth), GAL senses congestion globally using segmented injection queues to decide the directions to route in each dimension. It further load balances the network by routing in the selected directions adaptively. The use of global information enables GAL to achieve the performance (latency and throughput) of minimal adaptive routing on benign traffic patterns while performing as well as the best obliviously load-balanced routing algorithms (Valiant or GOAL) on adversarial traffic. We compare GAL to four other previously published routing algorithms on six figures of merit throughput on benign, adversarial and Hot-Spot traffic; average throughput of a sample of 1000 random permutations; latency at low injection load and stability post saturation. GAL is the only algorithm that has top performance in each category.