AHM-TF: An Adaptive Hop-by-Hop Multipath Transmission Framework

Conventional end-to-end multipath transmission selects complete paths at the source or ingress, limiting the ability of intermediate nodes to adapt forwarding decisions to locally observed network conditions. Hop-by-hop multipath transmission provides greater flexibility by allowing each node to select a next-hop option among multiple next hops, but it must jointly address forwarding-loop prevention, path efficiency, and adaptive traffic scheduling. This paper proposes the Adaptive Hop-by-Hop Multipath Transmission Framework (AHM-TF), which integrates multipath structure construction with hop-by-hop traffic scheduling. Its Weighted-Priority Loop Pruning (WPLP) procedure constructs destination-oriented, loop-safe next-hop sets while controlling structural detours. Over the resulting forwarding structure, Exponentially Weighted Hop-by-Hop Scheduling (EWHS) combines structural path cost with locally observed congestion pressure to update node-local traffic-splitting probabilities. For a fixed self-loop-free directed topology with positive structural link costs, WPLP prevents persistent forwarding loops and preserves destination reachability under the previous-hop exclusion rule. Experiments on five network topologies show that WPLP achieves higher path availability than several existing structure-construction methods while maintaining low mean relative path stretch. Its path availability is broadly comparable to LFID, whereas its mean relative path stretch is lower on all evaluated topologies for k≥3. Transmission experiments on the GEANT topology show that AHM-TF performs comparably to the strongest multipath baselines under low load and provides clearer benefits as contention increases. Under the medium load, AHM-TF improves aggregate goodput by 14.87%, reduces the block loss ratio by 12.43 percentage points, and reduces mean one-way block delay by 2.37% relative to ECN-WRR. Under the high load, it retains the best results on these three metrics. These improvements are accompanied by higher block delay variation and block reordering, indicating reduced temporal and ordering stability in block delivery.