Supporting adaptive routing in InfiniBand networks

InfiniBand is a new standard for communication between processing nodes and I/O devices as well as for interprocessor communication. The InfiniBand Architecture (IBA) supports distributed deterministic routing because forwarding tables store a single output port per destination ID. This prevents packets from using alternative paths when the requested output port is busy. Despite the fact that alternative paths could be selected at the source node to reach the same destination node, this is not effective enough to improve network performance. However using adaptive routing could help to circumvent the congested areas in the network, leading to an increment in performance. In this paper we propose a simple strategy to implement forwarding tables for IBA switches that supports adaptive routing while still maintaining compatibility with the IBA specifications. Adaptive routing can be individually enabled or disabled for each packet at the source node. The proposed strategy enables the use in IBA of any adaptive routing algorithm with an acyclic channel dependence graph. In this paper, we have taken advantage of the partial adaptivity provided by the well-known up*/down* routing algorithm. Evaluation results show that extending IBA switch capabilities with adaptive routing may noticeably increase network performance. In particular network throughput improvement can be, on average, as high as 46%.