A fully adaptive routing algorithm in torus switching fabrics

Recently, The k-ary n-cube direct interconnect network is considered to construct high performance switching fabric for Internet routers. Routing algorithm, which specify how packets can be sent along switching elements in network, are crucial for the high throughput and low latency of switching fabric. For maximum performance, the routing algorithm should exhibit the following important features: avoidance of deadlocks, ability to work well under varying traffic patterns and supporting QoS(Quality of Service). In this paper we present a recovery-based minimal routing algorithm to reach such targets. As with recover-based approach, the detection scheme has significant impact on the performance. A simple and efficient recover scheme is proposed in this paper, queue length and timer-based deadlock detection scheme (QTDS). Then, QTDS-based routing algorithm is proposed (QTDS-R). QTDS-R is applied on a three dimensional (3D) torus architecture. Changing the size of the fabric, the length of packets and the number of virtual channel, the performance of QTDS-R is measured in terms of throughput, average packets latency. The performance comparisons have been done on OPENT simulation tool. Numerical results show that this algorithm is better than other existed adaptive algorithms.