Legacy networks are often designed to operate with simple single-path routing, like shortest-path, which is known to be throughput suboptimal. On the other hand, previously proposed throughput optimal policies (i.e., backpressure) require every device in the network to make dynamic routing decisions. In this work, we study an overlay architecture for dynamic routing such that only a subset of devices (overlay nodes) need to make dynamic routing decisions. We determine the essential collection of nodes that must bifurcate traffic for achieving the maximum multicommodity network throughput. We apply our optimal node placement algorithm to several graphs and the results show that a small fraction of overlay nodes is sufficient for achieving maximum throughput. Finally, we propose a heuristic policy (OBP), which dynamically controls traffic bifurcations at overlay nodes. In all studied simulation scenarios, OBP not only achieves full throughput, but also reduces delay in comparison to the throughput optimal backpressure routing.
展开▼
