DTN Routing Schemes Overview

Updated 7 January 2026

DTN Routing Schemes are protocols designed for networks with intermittent connectivity, using a store-carry-forward approach to maximize delivery success.
They employ diverse methods—from epidemic flooding to prediction-based, probabilistic, and learning-driven algorithms—to balance delivery probability, delay, and resource use.
Practical implementations include geographic routing, social context awareness, and optimization frameworks, each tailored to specific mobility patterns and network demands.

Delay/Disruption-Tolerant Network (DTN) routing schemes enable data delivery in networks characterized by intermittent or sparse connectivity, highly dynamic topology, and potentially long link outages. DTNs operate under a store-carry-and-forward paradigm, allowing nodes to buffer messages and relay them opportunistically or according to predictions about future contacts. Over the last two decades, a variety of DTN routing strategies have been developed, addressing distinct scenarios—ranging from deep-space scheduled contacts to pocket-switched human networks—using epidemic algorithms, probabilistic or history-based heuristics, geographic/trajectory information, optimization frameworks, and learning-based decision models. Performance trade-offs among delivery probability, delay, overhead, and resource utilization drive both protocol design and real-world deployment criteria.

1. Fundamental DTN Routing Paradigms

DTN routing can be broadly categorized according to how nodes select forwarding opportunities and manage message replication:

Epidemic Routing: A baseline flooding approach wherein nodes exchange summary vectors during every contact, forwarding all non-duplicate messages to encountered peers. This saturates relay resources and yields maximum delivery probability but incurs prohibitively high buffer and bandwidth overhead (Kawano et al., 31 Dec 2025).
Probabilistic/History-Based Schemes: Protocols such as PRoPHET maintain encounter histories to estimate delivery predictabilities for each potential destination, forwarding only to nodes with higher predicted success, thereby reducing replication (Kawano et al., 31 Dec 2025).
Geographic and Trajectory-Aware Routing: Schemes leverage node position, velocity, and (sometimes) trajectory predictability. Examples include Vector, TBGR, GRONE, and Centroid/CenterMass Routing, which use geographic criteria to select or limit candidate relays (Cao et al., 2016, You et al., 2016, Rohrer, 2018).
Prediction-Driven Schemes: Strategies such as REAPER and ORION use predictive models of node contact time (e.g., autoregressive ARMA processes or semi-deterministic mobility) to choose contacts with high likelihood of near-term delivery and/or to optimize for delay and capacity jointly (Roy et al., 2013, Medjiah et al., 2012).
Social and Context-Aware Forwarding: Protocols like dLife compute time-varying social weights and centrality scores per time slot to route only along the strongest social ties, reducing wasteful replication (Moreira et al., 2014).
Hybrid Protocols (DTN-MANET): HYMAD dynamically splits the topology into diameter-bounded groups, running intragroup MANET routing while crossing groups via DTN-style store-carry-forward with controlled spray logic (Whitbeck et al., 2010, Whitbeck et al., 2010).
Optimization and Learning-Based Schemes: Linear programming, MDP/POMDP models, and Q-learning address resource allocation, uncertainty, and context adaptation, aiming for optimal or robust routing under constraints (Madoery et al., 2022, Stock et al., 25 Nov 2025, Yesuf et al., 2021).

2. Geographic and Trajectory-Aware Routing Schemes

Geographic DTN routing approaches exploit spatial information, often enhancing robustness and efficiency in mobile vehicular and urban deployment contexts:

Centroid and CenterMass Routing: Nodes maintain a running average ("centroid") of their position history, which smooths out GPS noise and resists transient location errors. Forwarding quotas are proportional to centroid distance; CenterMass additionally restricts replication to paths that reduce Euclidean distance to the destination centroid (Rohrer, 2018). These schemes exhibit sub-10 overhead ratios and achieve 60–80% delivery on urban benchmarks, with strong resilience to ±20m GPS error.
TBGR and TBHGR: TBGR applies threshold-based multi-copy forwarding in homogeneous mobility, estimating intersection time via node velocity and direction. TBHGR extends this for heterogeneous mobility, biasing copy distribution toward nodes with prior encounters in regions of interest. Their two-phase protocol overcomes local-maxima and maintains high delivery ratio under realistic group-centric urban mobility patterns (Cao et al., 2016).
GRONE: A purely one-hop, utility-driven replication protocol that combines distance and directionality to select the best relay among neighbors, complemented by a formal message redundancy removal mechanism. It achieves epidemic-level delivery ratios (~Epidemic) with 40–60% lower overhead, especially at low node speeds (You et al., 2016).

Prediction and historical patterns substantially influence DTN routing efficiency:

ARMA-Based Predictive Forwarding (ORION): Maintains ARMA contact models for each neighbor, enabling prediction of future contact times and durations. Forwarding is single-copy and exploits nearest-neighbor/geographic greedy heuristics, delivering lower delay and overhead than PRoPHET across a range of densities and speeds (Medjiah et al., 2012).
Social-Tie and Time-Slot Weighting (dLife): Leverages fine-grained contact weights and node "importance" based on observed slot durations, forwarding copies only along socially strong, temporally relevant paths. dLife matches or exceeds PRoPHET's delivery rates, but with 25x lower replication, validated in both simulation and real Bluetooth field trials (Moreira et al., 2014).
Context-Adaptive and Learning-Based Approaches (CARL-DTN): Dynamically fuses physical, social, and message context via fuzzy logic, with Q-learning estimating multi-hop delivery probability. Controlled replication based on real-time context and node density achieves higher delivery and lower overhead than both Spray-and-Wait and PRoPHET under most buffer/TTL regimes (Yesuf et al., 2021).

4. Multi-Copy, Spray-Based, and Hybrid Protocols

Controlled-copies and group-based routing address the classic tradeoff between delivery probability and replication overhead:

Spray-and-Wait (S&W) and Variants: Limits the number of copies per message, striking a balance between epidemic delivery and buffer usage. Binary spray, utility-guided handover (as in GRONE, TBGR), and group-level copy distribution (HYMAD) further optimize performance in different regimes (Whitbeck et al., 2010, You et al., 2016, Cao et al., 2016).
Hybrid DTN-MANET (HYMAD): Dynamically forms and maintains small-diameter groups using a fully-decentralized algorithm. Runs fast distance-vector routing inside connected groups and uses controlled inter-group DTN spraying between them, offering faster delivery and higher delivery ratio than classical S&W or Epidemic under moderate-to-high density and mobility (Whitbeck et al., 2010, Whitbeck et al., 2010).
Message Ferries and Cyclic Routing: Deployment of message ferries on cycles derived from planar MSQ (multi-scale quartered) networks, optimizing ferry rates (as service queues) for end-to-end delay and resilience to node or ferry failures. Local merging/splitting cycles allow recovery from failures with no global state (Hayashi, 2015).

5. Optimization, Scheduled Contacts, and Uncertainty-Aware Routing

Emerging applications in satellite networks and highly dynamic settings accelerate the use of global optimization and decision-theoretic tools:

Contact Graph Routing (CGR) and QoS-Aware Extensions: CGR computes K-shortest routes in a scheduled contact plan; more recent variants add “fewest-hops” selection (CGR-Hops) and multi-objective metrics (CGR-MO) to prioritize low-latency or tight-TTL traffic, minimizing downstream congestion and improving deadline compliance under heterogeneous loads (Madoery et al., 2022, Madoery et al., 2023).
Convex and Integer Linear Programming: Time-expanded ILP/LP models characterize performance upper bounds for delivery ratio and energy use, allowing benchmarking and weighting strategies for mixed latency objectives (Madoery et al., 2023, Madoery et al., 2022).
Markov Decision Processes (RUCoP, L-RUCoP, CGR-UCoP): Formulates routing as a multi-copy MDP over uncertain contact plans—each scheduled contact with explicit failure probability—yielding policies that optimize delivery under stochastic plan failures. Locally computable approximations deliver near-optimal performance; CGR-UCoP integrates these metrics into CGR for up to 25% higher delivery ratio over conventional CGR at high link-failure rates (Raverta et al., 2021).
Partially Observable MDPs (POMDP-DNF): Models routing under partial knowledge and correlated node failures (via CTMCs), using online Monte-Carlo Tree Search (POMCP). Demonstrates lower overhead and improved energy efficiency over both custody transfer and failure-oblivious CGR baselines, especially in environments with high node-failure rates (Stock et al., 25 Nov 2025).
Geometry-Accelerated Spanner Routing: In satellite constellations, maintaining geometric metric spanners (e.g., stable Delaunay subgraphs) enables sparse, low-stretch routes with far fewer recomputation events than full time-extended contact graphs. This yields nearly optimal delivery within 5% of baseline delays but with an order-of-magnitude fewer topology updates (Piekenbrock, 2024).

6. Mathematical Models: Epidemic, Incentive, and Analytical Frameworks

Analytical models inform both the design and the bounds of DTN routing schemes:

Epidemic and SIR/CISER Models: Compartmental ODE models (SIR, CISER) relate the dynamics of message spreading/flooding to biological epidemics, enabling quantitative analysis of delivery ratio, delay, and replication overhead. CISER introduces Exposed and Carrier states for buffer/energy-constrained relays, delivering higher efficiency than classic SIR abstraction (CC et al., 2016).
Incentive and Reward Mechanisms: Analytical frameworks for two-hop relay networks with heterogeneous nodes determine optimal relay rewards under various information regimes, deriving payments invariant to knowledge level, and introducing TTL counters to tune the delivery-memory tradeoff (El-Azouzi et al., 2017).
Genetic Improvement for Protocol Evolution: Grammar-based genetic programming automatically evolves protocol component sequencing, yielding protocol variants with up to twice baseline delivery probability and over two orders-of-magnitude lower overhead in urban network scenarios; evolved logics favor aggressive retransmission to best one-hop neighbors and longer buffer residency for greater delivery probability (Lorandi et al., 2021).

7. Evaluation, Applicability, and Deployment Considerations

Protocol efficacy depends on system parameters, mobility regime, and tradeoff priorities:

Overhead vs. Delivery: Pure epidemic routing maximizes delivery but is impractical for energy/memory-limited devices due to replication storms. Utility-, prediction-, and social-aware schemes achieve near-epidemic delivery with dramatically lower resource use, but rely on sufficiently accurate context modeling (Kawano et al., 31 Dec 2025, You et al., 2016, Moreira et al., 2014).
Buffer Constraints and Congestion: At small buffer sizes, controlled replication (e.g., CARL-DTN, HYMAD, GRONE) outperforms flooding and history-based schemes, sustaining delivery while suppressing buffer-induced drops (Yesuf et al., 2021, You et al., 2016).
Mobility Patterns and Heterogeneity: Homogeneous mobility favors geometric threshold and binary spray; TBHGR, social-aware, and learning-based methods yield additional gains in urban, clustered, or preference-driven scenarios (Cao et al., 2016, Moreira et al., 2014, Yesuf et al., 2021).
Extensibility and Integration: Many routing schemes can be integrated into existing DTN software (e.g., CGR, DIP), and extensions exploit advanced policy-based routing, social awareness, or learning modules to improve robustness without major changes to protocol stack (Neufeld, 2010, Whitbeck et al., 2010, Yesuf et al., 2021).

DTN routing remains an active research field, with ongoing efforts directed at robust performance under high uncertainty, efficient use of scheduled and opportunistic links, context and social modeling, policy and incentive-driven relaying, and scalable decision-making in large, dynamic topologies.