Source-Coded Online Algorithm for Multicast Subgraph Construction

TL;DR

This paper addresses scalable multicast delivery by proposing an online source-coded multicast framework that builds edge-disjoint or zone-aligned paths from a single source to multiple receivers without requiring in-network coding. The core method is a color-constrained BFS that incrementally integrates receivers, enforcing overlaps to be feasible for source coding and ensuring loop-free, unit-capacity flows. The authors demonstrate that the online construction achieves near-optimal throughput relative to network coding while avoiding the computational and deployment overhead of coding at intermediate nodes, and it outperforms traditional multicast trees in fairness and robustness. A complete source-side encoding strategy uses colored substreams and RS/Vandermonde-based coding to guarantee decodability, with extensive simulations on ER and WS topologies validating scalability and practical viability in next-generation networks.

Abstract

Multicast remains a fundamental mechanism for scalable content distribution, yet existing approaches face critical limitations. Traditional multicast trees suffer from path redundancy and inefficient utilization of network resources, while network coding, although capacity-achieving, incurs significant computational overhead and deployment challenges. In this paper, we introduce a source-coded multicast framework that exploits maximum-flow decomposition to construct multiple disjoint or partially overlapping paths from the source to all receivers. Our scheme incorporates a novel path redirection mechanism: when multiple overlaps occur between receiver flows, downstream paths are realigned at the first intersection, ensuring loop-free delivery while maximizing overall throughput. We develop algorithms for path construction, overlap detection, and iterative refinement of multicast subgraphs, and analyze their computational complexity. Through extensive evaluation on synthetic and real network topologies, we demonstrate that the proposed method consistently approaches the throughput of network coding with substantially lower encoding and decoding complexity, while significantly outperforming multicast tree constructions in terms of fairness, robustness to link failures, and delivery efficiency. These results position source-coded multicast as a practical and scalable solution for next-generation networks requiring high-throughput and adaptive group communication.

Figures (10)

• Figure 1: Illustrative Edmonds--Karp BFS: (a) layered expansion from $s$; (b) predecessor map; (c) reconstruction of the shortest $s{\to}r_i$ augmenting path.
• Figure 2: Residual graph before (a) and after (b) one augmentation. The arcs used by the shortest $s{\to}r_i$ path are removed, while unused and reverse arcs remain available for subsequent searches.
• Figure 3: Step-by-step online integration of three receivers: (a) disjoint paths for $r_1$ (blue, green); (b) integration of $r_2$ with one disjoint path (red) and one single overlap (green); (c) integration of $r_3$ with redirection at the first overlap (blue).
• Figure 4: Adversarial topology showing a strict loss under overlap-constrained integration. (a) Topology only. (b) Per-receiver EK: two edge-disjoint paths to every receiver. (c) Restricted EK: the first-touch overlap rule collapses one branch for $r_3$, reducing the multicast group throughput.
• Figure 5: Illustrative Watts--Strogatz example ($n{=}40$, $k{=}4$, $\beta{=}0.1$).