Enhancing DTN capacity
with Throwboxes
(work-in-progress)
Wenrui Zhao, Yang Chen,
Mostafa Ammar, Mark Corner,
Brian Levine, Ellen Zegura
Georgia Institute of Technology
University of Massachusetts Amherst
Delay Tolerant Networks (DTN)



DTNs: non-Internet-like networks
 Intermittent connectivity
 Large delays
 High loss rates
Examples of DTNs
 Tactical networks, disaster relief,
peacekeeping
 Interplanetary networks, rural village
networks
 Underwater acoustic networks
DTN features
 Store-Carry-and-forward
 Message switching
Capacity Limitation in DTNs

DTNs are intermittently connected


Potentially low throughput, large delay
Question: enough capacity for applications?

What if not?
Enhancing DTN Capacity



Use radios with longer range
Deploy a mesh network as infrastructure
Message ferrying
MF
M S
D

This presentation: Throwboxes
Our Work on MF/DTN








Ferry Route Design Problem [FTDCS 03]
MF with Mobile Nodes
[MobiHoc 04]
Efficient use of Multiple Ferries [INFOCOM 05]
The V3 Architecture: V2V Video Streaming [PerCom 05]
Ferry Election/Replacement [WCNC 05]
MF as a power-savings device [PerCom 05]
Multipoint Communication in DTNs/MF [WDTN 05, WCNC 06]
Power Management Schemes in DTNs/MF [SECON 05,
PerCom 05]

Road-side to Road-side relaying using moving vehicles
[WCNC 06]
Throwboxes



Basic idea: add new devices to enhance data
transfer capacity between nodes
Deploy throwboxes to relay data between
mobile nodes
Throwboxes are:



small, inexpensive, possibly dispensable, batterypowered wireless devices
Some processing and storage capability
Easy to deploy and replenish
Throwboxes
Processor
Intel PXA255 400MHz
Memory
64MB SDRAM
32MB Flash
Power
consumption
< 500mA
Size
3.5’’ x 2.5’’
Weight
47g
Example:
DTN w/out Throwboxes
Example:
DTN w/ Throwboxes
UMassDiesel DTN Example
25000
throw-box
Bus 30
Bus 31
Bus 34
Bus 35
Bus 36
Bus 38
Bus 39
Bus 39e
Bus 45
20000
Y
15000
10000
5000
w/out
TB
w/ TB
Total contact
duration (sec)
631
11927
Effective capacity
(Kbps)
3.5
66.3
63012
3120
Delay (sec)
0
0
5000
10000
15000
20000
25000
X


Data transmission between bus 38 and bus 45
A single throwbox achieves an improvement factor of 19 for
both capacity and delay
Main Question

How to best deploy



‘s
Where?
How to route through them?
When? -- Later work
Throwbox Deployment &
Routing Framework

Objective: throughput enhancement


Important to deliver data
May improve delay too


Deployment issue


Where to place throw-boxes?
Routing issue

How data are forwarded?





Contact-oblivious
Contact-based
Traffic and Contact
based
Single path routing
Multi-path routing
Epidemic routing
Network Model


DTN consists of mobile nodes
Relative traffic demand between nodes bij
b
ij
1
i, j


Total throughput λ
Given inherent capacity (w/out TBs) as a
function of:
 Contacts – dictated by mobility patterns
 Data rate
Throwbox Assumptions



Sufficient energy supplies
No interaction between throwboxes
Deployed to a given set of potential locations


Center of Grid Cells
Deployment Vector (0/1 vector)
Throwbox Deployment &
Routing Framework
Deployment approach
Traffic &
Contact
based
Contact
based
Contact
oblivious
Random or Regular Deployment
Multi-path
routing
Single path
routing
Epidemic
routing
Routing
approach
Throwbox Deployment &
Routing Framework
Deployment approach
Traffic &
Contact
based
Contact
based
Contact
oblivious
Random or Regular Deployment
Multi-path
routing
Single path
routing
Epidemic
routing
Routing
approach
Multi-Path Routing –
Traffic and Contact-Aware Deployment

Need to determine



Deployment locations of throwboxes
Routing paths and traffic load on each path
Performance objective

Given m throwboxes, maximize total throughput λ
such that traffic load λbij is supported from node i
to j
Multi-Path Routing –
Traffic and Contact-Aware Deployment

Formulated as an 0/1 linear programming
problem



Throwbox deployed at location  1
Solution also gives optimal flow vector describing
use of multiple paths
NP-hard to solve optimally
Greedy Heuristic

Deploy throwboxes one by one
(1) for i=1 to m do
(2)
find location L that maximizes λ
(3)
deploy a throwbox at location L
(4) end
(5) compute routing

Given throwbox locations, (2) is a concurrent flow
problem

Solved by network flow techniques or linear programming
tools
Throwbox Deployment &
Routing Framework
Deployment approach
Traffic &
Contact
based
Contact
based
Contact
oblivious
Random or Regular Deployment
Multi-path
routing
Single path
routing
Epidemic
routing
Routing
approach
Multi-Path Routing –
Contact-Based Deployment

Throwbox deployment is based on contact
information, but not traffic information



Benefits varying traffic patterns
May not be optimal for specific traffic
Maximize


Absolute contact enhancement
 Maximize absolute enhancement of contact between nodes
Relative contact enhancement
 Maximize relative enhancement of contact between nodes
Throwbox Deployment &
Routing Framework
Deployment approach
Traffic &
Contact
based
Contact
based
Contact
oblivious
Random or Regular Deployment
Multi-path
routing
Single path
routing
Epidemic
routing
Routing
approach
Single Path Routing

Single path routing


Data for a S-D pair follow a single path
Adapt greedy algorithm for multi-path routing by
enforcing the “single path” requirement
Throwbox Deployment &
Routing Framework
Deployment approach
Traffic &
Contact
based
Contact
based
Contact
oblivious
Random or Regular Deployment
Multi-path
routing
Single path
routing
Epidemic
routing
Routing
approach
Epidemic Routing

Epidemic routing (ER)


Difficult to characterize traffic load among nodes
because of flooding
ER exploits all paths to propagate data
 Multi-path heuristic
 Proportional allocation heuristic
Performance Evaluation
traffic demand
node mobility
deployment/routing
computation

throwbox locations
routing path/load
ns simulation
Objectives

Utility of throwboxes in performance enhancement

Performance impact of various routing and deployment approaches
Simulation Settings

Node mobility models




Predictable/constrained: UMass model based on measured
bus trace
Random/unconstrained: Random waypoint model
Random/constrained: Manhattan model
Simulation Parameters




9 nodes in a 25Km x 25 Km area
802.11 MAC, radio range: 250m, bandwidth: 1Mbps
20 source-destination pairs, message size is 1500 bytes,
Poisson message arrival with same data rate
FIFO buffer, buffer size is 50000 messages
Delivery Ratio vs. Number of
Throwboxes
0.8
T &C Aware
AbsoluteContact
RelativeContact
Random
Grid
Multi-path
routing
Message delivery ratio
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0
1
2
3
4
5
Number of throw-boxes
6
7
8
Delivery Ratio vs. Number of
Throwboxes
0.45
Single path
routing
Message delivery ratio
0.4
0.35
0.3
0.25
0.2
0.15
T & C Aware
0.1
AbsoluteContact
RelativeContact
Random
Grid
0.05
0
0
1
2
3
4
5
Number of throw-boxes
6
7
8
Delivery Ratio vs. Number of
Throwboxes
0.35
Epidemic
routing
Message delivery ratio
0.3
0.25
0.2
0.15
MultiPath
Proportional
AbsoluteContact
RelativeContact
Random
Grid
0.1
0.05
0
0
1
2
3
4
5
Number of throw-boxes
6
7
8
Delay vs. Number of Throwboxes
(High Traffic Load)
14000
Multi-path
routing
Message delay (second)
12000
10000
8000
6000
4000
T&C
AbsoluteContact
RelativeContact
Random
Grid
2000
0
0
1
2
3
4
5
6
Number of throw-boxes
7
8
Delay vs. Number of Throwboxes
(Low Traffic Load)
Multi-path
routing
Message delay (second)
6000
5000
4000
3000
2000
T&C
AbsoluteContact
RelativeContact
Random
Grid
1000
0
0
1
2
3
4
5
Number of throw-boxes
6
7
8
Summary of Simulation Results
UMass
mobility
Throughput
improvement
Delay improvement
(high traffic load)
Manhattan
mobility
Delay improvement
(low traffic load)
RWP
mobility
Multi-path
routing
Single path
routing
Epidemic
routing
Summary of Simulation Results (2)
High
T&C/
Contact based
T&C
Throughput
improvement
Contact based
T & C/
Contact based
Low
Contact
oblivious
Multi-path
routing
Contact
oblivious
Single path
routing
Contact
oblivious
Epidemic
routing
Routing
approach
Summary

Study the use of throwboxes for capacity enhancement in mobile
DTNs

Develop algorithms for throwbox deployment and routing
 Routing: multi-path, single path, epidemic
 Deployment: traffic and contact, contact-based, contact-oblivious

Evaluate the utility of throwboxes and various routing/deployment
approaches
 Throwboxes are effective in improving throughput and delay,
especially for multi-path routing and predictable node mobility
Questions?
Message Ferrying
MF
S
S
M
MF
D
M
D