Recently , I have seen the source code of the up*/down* routing algorithm. It still uses the BFS to generate the spanning tree, whoes performance is inferior to the improved one which uses DFS to generate the spanning tree.
DFS have been proposed in paper "New Methodology to Compute Deadlock-Free Routing Tables for Irregular Networks", which can increase throughput and reduce latency. So I think that It's a better idea to use DFS to generate the spanning tree in up*/down* routing algorithm. Why not use DFS in up*/down* routing algorithm ? Is it hard to implement or some other reason ?
_______________________________________________ general mailing list [email protected] http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
