Answers are grounded in this paper and cite their sources.
In disaster recovery, deep-space missions, and other intermittently connected deployments, delay-tolerant networks deliver bundles by replicating them at every encounter, and copies of a bundle that has already reached its destination continue to consume scarce forwarding capacity. This post-delivery replication is the principal avoidable overhead in epidemic delay-tolerant networking (DTN). Oracle-based proposals address it by instructing nodes to delete copies on delivery, but this approach requires a synchronised, network-wide state update that intermittently connected nodes cannot sustain. We present P4DTN, a Mininet/BMv2 emulation platform that realises in-network bundle suppression using P4. Switches recognise an already-delivered bundle from a compact identifier and drop it at fabric ingress, before the multicast fan-out that would otherwise replicate it across every outbound link. An Oracle family installs these drop rules on each confirmed delivery, through three nested variants that differ only in how actively they manage host injection. The result is a suppression mechanism enforced at the gateway switches of an infrastructure-assisted DTN. Hosts are no longer required to take part in cleanup. Coordination is limited to one rule installation per gateway per delivery, issued by a central Oracle over a reliable control channel to a small, fixed set of switches rather than to every intermittently connected node. Experiments on an emulated P4 fabric show that every Oracle variant raises the offered delivery ratio from a No-Oracle baseline median of 59% to between 76% and 79%, reduces total traffic volume by 28%–32%, and removes approximately 87% of post-delivery duplicates. Under this many-to-one epidemic workload, the improvement comes from higher forwarding success rather than heavier injection, so delivery gain and load reduction move together rather than trading off. The same pattern holds on a much larger network, where offered delivery rises from a baseline of 39% to 56% and total traffic volume falls by approximately 15%.
No takes yet. Share an insight, caveat, or question.
Topu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: