A Deadlock Recovery Based Adaptive Routing Algorithm for Multi-dimensional Switching Fabric

Scalable switching fabrics can be done on implementing high performance routers by employing multi-dimensional packet switching fabrics. The internal routing algorithm in the switching fabric is a key technology. This paper proposes a new distributed deadlock detection strategy, which combines with queue length on each node without other information required except local information. Based on this technology, a fully adaptive routing algorithmQDAR(Queue length based Deadlock recovery Adaptive Routing) have been designed. The performance is assessed on 3-dimensional torus architecture. Effect of the buffer length, the number of virtual channels and variable traffic types has been analyzed. Performance evaluation through comparing with other routing algorithms has been done.