Tunable Multi-Objective Tree Synthesis for Application-Layer Multicast

Abstract

Application-layer multicast over wide-area networks must reconcile competing goals: paths should be short while links should offer high capacity and remain stable under variability. This paper presents a method that preserves multi-objective structure and exposes an explicit control of source-to-receiver latency. Edge preferences are derived from round-trip time (RTT) and available bandwidth (AvB) using NSGA-II and a seed tree is obtained by a minimum-spanning construction over the rank matrix. Path length is then constrained by a LAST-style adjustment evaluated in RTT, which guarantees that the distance to every receiver does not exceed a chosen factor α of the RTT shortest path. An equivalent additive budget L is supported for operational use. The approach is implemented with RMDT transport and evaluated on a five-region AWS testbed with active RTT/AvB probing and controlled cross-traffic. The NSGA-II–seeded trees consistently outperform singlemetric baselines, and the latency constraint modifies only those segments that would otherwise incur excessive delay. The result is a monotone, single-parameter knob that lets operators satisfy latency requirements while retaining the performance benefits of multi-objective structure.