G3M: A heuristic for low-latency multicast back-pressure forwarding
Stephen Dabideen, Stephen Zabele, Christophe J. Merlin, Laura Poplawski, Gregory Lauer · 2019
It has been previously proven that optimal solutions to mulitcast routing and scheduling can be achieved using back-pressure together with network coding. However, this optimality comes at the cost of increased latency and complexity, which cannot be afforded in tactical military operations. In this work, we present the Greatest Group Gradient for Multicast over back-pressure (G3M), a heuristic that greatly reduces complexity, state and latency when compared to solutions that use network coding, while approaching the same optimal solutions. Unlike most prior approaches, G3M does not rely on the construction of explicit multicast trees. Instead, it uses only queue information from the 1-hop neighborhood to make distributed forwarding decisions, that are globally optimal. Additionally, we present updates to the admission control algorithm that promotes fairness between multicast flows with different characteristics. With this in place G3M can use the same forwarding algorithm can be used with both unicast and multicast traffic. We also introduce an algorithmic enhancement to back-pressure forwarding, which we call opportunistic forwarding, to reduce the replication of multicast packets, allowing the heuristic to operate closer to the optimal solution during transient network conditions. Our results show that the G3M heuristic converges to the provable optimal solution.