68  Delay-Tolerant Networks for IoT

In 60 Seconds

Delay-Tolerant Networks use store-carry-forward to deliver data when end-to-end connectivity never exists simultaneously. Epidemic routing achieves highest delivery (90%+) but floods the network with O(N^2) copies; Spray-and-Wait limits copies to L (typically 6-12) achieving 80-85% delivery with 10x less overhead; PRoPHET uses encounter history for 75-85% delivery with intelligent forwarding. Choose based on your buffer/bandwidth vs delivery requirement tradeoff.

68.1 Learning Objectives

By the end of this chapter, you will be able to:

• Apply DTN Concepts: Select Delay-Tolerant Networking for appropriate intermittent connectivity scenarios
• Compare Routing Protocols: Evaluate Epidemic, Spray-and-Wait, and PRoPHET routing trade-offs using delivery rate, overhead, and latency metrics
• Design Store-Carry-Forward Systems: Architect data collection pipelines for disconnected environments
• Calculate Delivery Probability: Compute PRoPHET predictability values and Bayesian delivery estimates

Key Concepts

• Core Concept: Fundamental principle underlying Delay-Tolerant Networks for IoT — understanding this enables all downstream design decisions
• Key Metric: Primary quantitative measure for evaluating Delay-Tolerant Networks for IoT performance in real deployments
• Trade-off: Central tension in Delay-Tolerant Networks for IoT design — optimizing one parameter typically degrades another
• Protocol/Algorithm: Standard approach or algorithm most commonly used in Delay-Tolerant Networks for IoT implementations
• Deployment Consideration: Practical factor that must be addressed when deploying Delay-Tolerant Networks for IoT in production
• Common Pattern: Recurring design pattern in Delay-Tolerant Networks for IoT that solves the most frequent implementation challenges
• Performance Benchmark: Reference values for Delay-Tolerant Networks for IoT performance metrics that indicate healthy vs. problematic operation

68.2 Prerequisites

Before diving into this chapter, you should be familiar with:

• Human-Centric Sensing: Roles and Paradigms: Understanding of human mobility patterns and how they enable data collection in disconnected scenarios
• Wireless Sensor Networks: Knowledge of WSN topologies and routing fundamentals provides context for understanding when traditional routing fails
• Networking Basics: Familiarity with networking fundamentals is essential for grasping how DTN differs from traditional IP networking

For Beginners: Delay-Tolerant Networks Explained

The Disconnection Problem:

Traditional networks assume devices can always communicate. But what if you’re tracking wildlife in a remote forest with no cell signal? Delay-Tolerant Networks solve this using “store-carry-forward”:

1. Store: A zebra collar sensor stores GPS data locally
2. Carry: The zebra physically moves through the savanna carrying the data
3. Forward: When the zebra visits a watering hole with a base station, it uploads all stored data

Real-World Analogy: Think of it like delivering a package via multiple delivery trucks rather than calling on the phone. Each truck carries the package part of the way until it reaches its destination - slow but reliable when phone lines don’t exist.

Where DTN is Used:

• Wildlife tracking: Animal collars in remote areas with no network
• Space communication: Mars rovers sending data back to Earth
• Disaster recovery: Communication after earthquakes destroy infrastructure
• Rural connectivity: Villages connected by buses carrying data
• Ocean monitoring: Sensors on buoys communicating via passing ships

Term	Simple Explanation
DTN (Delay-Tolerant Network)	Networks that work even when connections are intermittent - like rural mail delivery
Store-Carry-Forward	Save data, physically move it, then send when possible - like carrying a USB drive
Epidemic Routing	Spread data copies like a disease - give to everyone you meet to ensure delivery
Opportunistic Contact	When two devices come within communication range unexpectedly
Data MULE	Mobile node that physically transports data between disconnected network segments

Why This Matters for IoT:

Traditional sensor networks fail in scenarios like ocean monitoring, rural areas, or wildlife tracking. DTN enables data collection where infrastructure doesn’t exist - perfect for developing regions, disaster recovery, and environmental research where continuous connectivity is impossible.

Minimum Viable Understanding (MVU)

If you only learn three things from this chapter:

1. DTNs use store-carry-forward – when no end-to-end path exists, nodes store data locally, physically move (carry), and forward when they encounter another node or destination
2. Three routing protocols trade delivery vs overhead – Epidemic (90-95% delivery, ~100x copies), Spray-and-Wait (80-85% delivery, 6-10x copies), and PRoPHET (70-80% delivery, 3-5x copies using encounter history)
3. Real systems prove DTN works – ZebraNet achieved 85% delivery for wildlife tracking with 7-hour latency, and DakNet achieved 99.9% delivery for rural email at 90% less cost than telephone lines

Sensor Squad: The Message Relay Race!

Sammy the Sensor is stuck on a zebra in the African savanna with no phone signal. How does he send his data home?

Lila the Listener explains: “Imagine you have a letter to deliver, but there’s no post office. Instead, you give copies of your letter to everyone you meet. Some of those people will meet other people, who will meet other people… and eventually, someone will walk past the mailbox!”

Max the Messenger describes three delivery strategies:

• The Copy-Everything Plan (Epidemic): “Give a copy to EVERYONE you meet. Your letter WILL arrive, but you’ll use up all your paper making copies!” (95% delivery, lots of waste)
• The Limited-Copies Plan (Spray-and-Wait): “Make only 6 copies and hand them to the first 6 people. Then everyone just waits to bump into the mailbox.” (80% delivery, much less waste)
• The Smart Plan (PRoPHET): “Only give your letter to people who regularly walk near the mailbox. Your friend who works near the post office is a MUCH better carrier than a random stranger!” (70% delivery, very little waste)

Bella the Battery asks: “But isn’t waiting hours for delivery too slow?”

Sammy says: “For my zebra GPS data, waiting a few hours is fine! Scientists don’t need to know where I am RIGHT NOW – they just want to study the zebra’s patterns over time. DTN is perfect for patient data!”

68.3 Delay Tolerant Networks (DTNs)

Delay Tolerant Networks represent a networking paradigm designed for environments where traditional assumptions (end-to-end connectivity, low latency) do not hold.

68.3.1 DTN Characteristics

1. Intermittent Connectivity

• No guaranteed end-to-end path at any instant
• Communication opportunities arise and disappear
• Links may be available only when nodes meet

2. Long Delays

• Message delivery can take hours or days
• Propagation delays can be significant (especially in space)
• Store-and-forward over long time scales

3. Asymmetric Data Rates

• Forward and reverse links may have different rates
• Opportunistic contacts have varying quality
• Channel conditions vary over time

4. Resource Constraints

• Limited buffer space
• Energy constraints
• Bandwidth scarcity

Figure 68.1: Diagram showing IoT architecture components and their relationships with data flow and processing hierarchy

Delay-Tolerant Network (DTN) store-carry-forward mechanism diagram: Shows non-traditional routing through physical mobility: Node A has message M with no direct network path to Node B (indicated by dashed line showing disconnection), Node A stores message in local buffer waiting for opportunity, mobile relay node physically moves through network carrying stored data in buffer, opportunistic contact occurs when mobile node comes within communication range of Node B, message forwarded from mobile node to Node B upon contact, Node B stores message in local buffer for custody transfer, eventually message delivered to final destination despite intermittent connectivity and long delays enabling communication in challenged networks (wildlife tracking, disaster scenarios, rural IoT deployments).

68.3.2 DTN in Wireless Sensor Networks

Traditional WSNs assume that sensor nodes can route data to the base station through multi-hop paths. However, in many scenarios, this assumption doesn’t hold:

Sparse Deployment:

• Nodes too far apart for multi-hop connectivity
• Monitoring large areas with limited nodes

Harsh Environments:

• Obstacles blocking communication
• Intermittent power availability
• Extreme conditions causing failures

Mobile Scenarios:

• Mobile nodes move in and out of range
• Wildlife tracking with animal-borne sensors
• Vehicular sensor networks

Solution: Delay Tolerant WSNs (DT-WSNs)

• Embrace mobility and intermittent connectivity
• Use data MULEs or mobile sinks
• Store-and-forward with long delays acceptable

Figure 68.2: Four DTN application scenarios: Wildlife tracking uses animal-borne sensors that store GPS data locally until animals visit watering holes with base stations. Disaster recovery enables communication when infrastructure is destroyed through relief workers carrying devices that mesh when in range. Rural connectivity connects isolated villages via buses carrying data packets to town internet gateways. Ocean monitoring deploys buoy sensors that transfer data to passing ships which carry it to port for satellite uplink.

68.3.3 DTN Routing Protocols

Since DTN links are opportunistic and intermittent, routing becomes fundamentally different from traditional networks.

1. Epidemic Routing {#arch-wsn-part-epidemic}

Concept: When two nodes meet, they exchange all messages they carry (like spreading an epidemic).

Algorithm:

When node A encounters node B:
  For each message M in A's buffer:
    If B doesn't have M:
      A transmits copy of M to B

  For each message M in B's buffer:
    If A doesn't have M:
      B transmits copy of M to A

Advantages:

• Maximizes delivery probability
• Fast message propagation
• Robust to node failures

Disadvantages:

• High resource consumption (buffer, bandwidth, energy)
• Many redundant copies
• Can lead to network congestion

2. Spray and Wait {#arch-wsn-part-spray-wait}

Concept: Limit message replication to reduce overhead.

Phases:

1. Spray Phase: Distribute L copies of message to first L distinct nodes
2. Wait Phase: Nodes carrying a copy perform direct transmission only (wait for destination)

Variants:

• Binary Spray and Wait: Each node forwards half its copies and keeps half
• Source Spray and Wait: Source distributes all L copies

Advantages:

• Lower overhead than Epidemic
• Better resource utilization
• Still maintains good delivery probability

3. PRoPHET (Probabilistic Routing Protocol using History of Encounters and Transitivity) {#arch-wsn-part-prophet}

Concept: Use contact history to estimate delivery predictability.

Delivery Predictability P(A, B):

• High if A frequently encounters B
• Decays over time if no encounters
• Transitive: if A encounters B often, and B encounters C often, then P(A, C) should increase

Update Equations:

Encounter: P(A, B) = P_old(A, B) + (1 - P_old(A, B)) * P_init
Aging: P(A, B) = P_old(A, B) * gamma^k
Transitivity: P(A, C) = P_old(A, C) + (1 - P_old(A, C)) * P(A, B) * P(B, C) * beta

Where: - P_init is in (0, 1): initialization constant - gamma is in (0, 1): aging constant - beta is in (0, 1): scaling constant for transitivity - k: number of time units elapsed

Putting Numbers to It

PRoPHET Delivery Predictability Calculation: Node A (commuter) encounters node B (bus) every morning at 8am with \(P_{init} = 0.75\). After 3 consecutive encounters:

\[P(A,B) = 0 + (1-0) \times 0.75 = 0.75\] (first encounter) \[P(A,B) = 0.75 + (1-0.75) \times 0.75 = 0.9375\] (second encounter) \[P(A,B) = 0.9375 + (1-0.9375) \times 0.75 = 0.984\] (third encounter)

With no encounters for 5 days (\(\gamma = 0.98\), k = 5):

\[P(A,B) = 0.984 \times 0.98^5 = 0.984 \times 0.904 = 0.889\]

Bus B encounters gateway C daily with \(P(B,C) = 0.95\). Transitivity (\(\beta = 0.25\)):

\[P(A,C) = 0 + (1-0) \times 0.889 \times 0.95 \times 0.25 = 0.211\]

Node A forwards a message to bus B because \(P(B,C) = 0.95 > P(A,C) = 0.211\). The 0.95 delivery probability through B is 4.5× better than A’s direct probability, making B an ideal carrier despite the intermediate hop.

Figure 68.3: Diagram showing IoT architecture components and their relationships with data flow and processing hierarchy

DTN routing protocol comparison: Epidemic routing floods copies to every encountered node achieving 90-95% delivery with approximately 100x overhead (use for critical messages), Spray-and-Wait limits to L=6 copies balancing 80-85% delivery with 6x overhead (recommended for most IoT scenarios), PRoPHET uses encounter history forwarding only to nodes with higher delivery probability achieving 70-80% delivery with 3-5x overhead (optimal for predictable mobility patterns like bus routes or commuter networks).

Mid-Chapter Check: Test your understanding of DTN routing protocols before exploring the selection guide:

68.3.4 Routing Protocol Selection Guide

Criteria	Epidemic	Spray-and-Wait	PRoPHET
Delivery Rate	90-95%	80-85%	70-80%
Latency	Lowest	Medium	Medium-High
Overhead	~100x copies	6-10x copies	3-5x copies
Buffer Needed	Very High	Medium	Low-Medium
Energy Cost	Highest	Medium	Lowest
Best For	Critical alerts	General IoT	Predictable routes
Mobility Pattern	Any/Random	Any	Regular/Scheduled

Selection Guidelines:

1. Use Epidemic when:
   • Message is critical (disaster alerts, medical emergencies)
   • Buffer and energy are not constraints
   • Network is small (<50 nodes)
   • Must guarantee delivery
2. Use Spray-and-Wait when:
   • Balancing delivery vs. overhead
   • General IoT data collection
   • Medium-sized networks
   • Unknown mobility patterns
3. Use PRoPHET when:
   • Nodes have predictable movement (buses, commuters)
   • Resources are constrained
   • Can tolerate lower delivery rates
   • Contact patterns are stable

68.3.5 Case Study: ZebraNet

ZebraNet (Princeton, 2004) pioneered DTN for wildlife tracking:

Deployment:

• GPS collars on zebras in Kenya
• Base station at watering hole
• No cellular coverage in reserve

DTN Implementation:

• Store: Collar logs GPS every 3 minutes (288 points/day)
• Carry: Zebra movement through savanna
• Forward: When zebra visits watering hole, peer-to-peer transfer to base station

Routing Protocol: Modified Epidemic - Exchange data when two zebras encounter each other - Eventually, one zebra visits base station and uploads collective data

Results:

• 85% data delivery rate
• Average latency: 7 hours (acceptable for research)
• Discovered unexpected migration patterns
• Battery life: 2+ years with duty cycling

Key Insights:

• Animal behavior creates natural data mules
• Social animals (herding) enable peer-to-peer data spreading
• Watering holes as network hubs exploit predictable behavior

68.3.6 Case Study: DakNet

DakNet (MIT Media Lab) demonstrated DTN for rural connectivity:

Problem:

• Remote Indian villages without internet
• Telephone lines too expensive to deploy
• Villages separated by 5-10km

Solution:

• Mount wireless access points on buses
• Villages have local kiosks with Wi-Fi
• Buses physically carry data between villages

Implementation:

1. Villager composes email at kiosk
2. Email stored on kiosk computer
3. Bus drives through village, auto-connects to kiosk
4. Email transferred to bus storage
5. Bus reaches town with internet gateway
6. Email uploaded to internet
7. Replies downloaded to bus
8. Return trip delivers replies to village kiosks

Results:

• 99.9% message delivery rate
• Average latency: 4-8 hours (one bus trip)
• Cost: 90% less than telephone deployment
• Services: Email, government forms, telemedicine

---

68.4 Knowledge Check

Test your understanding of Delay Tolerant Networks.

Quiz: Delay Tolerant Networks

Match: DTN Concepts and Applications

Order: DTN Store-Carry-Forward Message Delivery

Worked Example: Rural Health Clinic Data Relay via Bus-DTN

Scenario: A public health programme in rural Malawi needs to collect daily malaria rapid-diagnostic-test (RDT) results from 12 clinics spread across a 60 km x 40 km district. No clinic has internet; the nearest fibre connection is at the district hospital. Two public bus routes pass through most clinics once per day.

Given:

• 12 clinics, each generating 50-200 RDT results per day (average 2 KB per record after compression = 100-400 KB/day per clinic)
• District hospital has fibre internet and the national disease-surveillance dashboard
• Bus Route A visits clinics 1-7 (morning loop, returns to hospital by 14:00)
• Bus Route B visits clinics 5-12 (afternoon loop, returns to hospital by 18:00)
• Clinics 5-7 are on both routes (overlap zone)
• Each clinic has a Raspberry Pi kiosk with Wi-Fi; buses have Pi + 4G dongle
• Wi-Fi transfer rate at stop: 2 Mbps effective; average bus stop duration: 90 seconds
• Target: 99% data delivery within 24 hours

Step 1 – Capacity check per stop: Transfer window = 90 seconds at 2 Mbps = 22.5 MB capacity per stop. Maximum clinic payload = 400 KB/day. Utilisation per stop = 400 KB / 22,500 KB = 1.8%. Massive headroom – even if 50 clinics shared one route, capacity would suffice.

Step 2 – Delivery probability analysis: Using Spray-and-Wait with L = 2 copies (one per bus route for overlap clinics 5-7):

Clinic group	Copies	Routes available	Delivery probability (single day)
1-4 (Route A only)	1	1	98% (bus runs except holidays)
8-12 (Route B only)	1	1	98%
5-7 (both routes)	2	2	99.96% (1 - 0.02 x 0.02)

Overall single-day delivery: (7 x 0.98 + 2 x 0.98 + 3 x 0.9996) / 12 = 98.3%. Over a rolling 48-hour window (two bus cycles), probability of at least one delivery = 1 - (0.02)^2 = 99.96% for single-route clinics.

Step 3 – Latency budget: Bus A departs hospital 07:00, visits clinic 1 at 07:45 … clinic 7 at 11:30, returns 14:00. Worst-case latency = data generated at 14:01 (after bus departure), collected next day at 11:30 = 21.5 hours. Average latency = 8 hours.

Step 4 – Cost comparison:

Approach	Monthly cost	Reliability
Cellular modem per clinic (if available)	$720 (12 x $60/month data plan)	85% (poor rural coverage)
VSAT satellite per clinic	$3,600 (12 x $300/month)	99%
Bus-DTN (this design)	$120 (2 Pis + electricity)	99% (48-hour window)

5-year total: Bus-DTN $7,200 vs cellular $43,200 vs VSAT $216,000.

Result: Bus-DTN achieves 99%+ delivery reliability at 83% lower cost than cellular and 97% lower cost than satellite. The 8-hour average latency is acceptable for daily disease surveillance reporting.

Key Insight: DTN works best when latency tolerance is measured in hours, not seconds. Public transport creates free, predictable mobility – the “carrier” cost is zero because buses already run on schedule. The overlap zone (clinics 5-7) on both routes provides natural redundancy without any additional infrastructure.

🏷️ Label the Diagram

Code Challenge

68.5 Summary

This chapter covered Delay-Tolerant Networks for IoT applications:

• DTN Characteristics: Designed for intermittent connectivity, long delays, asymmetric data rates, and resource-constrained environments where traditional end-to-end assumptions fail
• Store-Carry-Forward: Fundamental mechanism where data is stored locally, physically carried through node movement, and forwarded when encountering destination or relay nodes
• Epidemic Routing: Maximizes delivery probability (90-95%) by replicating messages to every encountered node, at cost of approximately 100x overhead - use only for critical messages
• Spray-and-Wait: Balances delivery (80-85%) and overhead (6x) by limiting to L copies then waiting for direct delivery - recommended for most IoT scenarios
• PRoPHET Routing: Uses encounter history for probabilistic forwarding, achieving 70-80% delivery with 3-5x overhead - optimal for predictable mobility patterns like bus routes
• Real-World Applications: ZebraNet wildlife tracking (85% delivery, 7-hour latency), DakNet rural connectivity (99.9% delivery, 4-8 hour latency), demonstrating DTN viability for challenged networks

Visual Reference Gallery

68.5.1 DTN Store-and-Forward

DTN Architecture

Store-and-forward mechanism enabling communication through disconnected networks.

68.5.2 Mobile Sink Routing

Mobile Sink

Mobile sink routing strategies for energy-efficient data collection.

68.6 What’s Next

Topic	Chapter	Description
Human-Centric Sensing	Human-Centric Sensing	Human roles as targets, operators, and data sources in sensing systems
Participatory Sensing	Participatory Sensing	Platform architecture and crowdsourced data collection applications
WSN Mobile vs Stationary	WSN Stationary vs Mobile	Mobile sinks and data MULEs for energy-efficient data collection

Cross-Hub Connections: Explore DTN Topics

Hands-On Simulations:

• Simulations Hub features DTN routing simulators (THE ONE simulator) where you can experiment with Epidemic, Spray-and-Wait, and PRoPHET protocols, comparing delivery rates and overhead in different mobility scenarios

Test Your Knowledge:

• Quizzes Hub contains DTN quizzes covering routing algorithms, store-carry-forward mechanisms, and protocol selection with detailed explanations

Video Learning:

• Videos Hub offers curated videos on delay-tolerant networking in wildlife tracking, disaster recovery, and rural connectivity scenarios

Common Pitfalls:

• Knowledge Gaps Hub explains misconceptions about DTN overhead (epidemic routing creates 100x copies), delivery guarantees, and when to use each routing protocol

Related Chapters

Human-Centric Sensing:

• Human-Centric Sensing: Roles and Paradigms - Human roles in sensing systems
• Participatory Sensing: Platforms and Applications - Crowdsourced data collection

WSN Fundamentals:

• Wireless Sensor Networks - WSN architecture fundamentals
• WSN Stationary vs Mobile: Fundamentals - Mobile sinks and data MULEs
• WSN Overview: Fundamentals - Network basics

Routing & Protocols:

• Sensor Network Routing - Data-centric routing approaches
• RPL Routing - Low-power routing
• MQTT - Lightweight messaging protocol

Learning Hubs:

• Simulations Hub - DTN simulation tools
• Videos Hub - Tutorial videos