Networking Research in the International Technology Alliance -

radiographerfictionData Management

Oct 31, 2013 (3 years and 7 months ago)

68 views

© 2007 IBM Corporation

IBM T J Watson Research Center

Slide
1

Invited talk at KAIST, 4/30/2007

Networking Research in the International Technology
Alliance

-

Topology control and data dissemination in wireless
networks

Kang
-
Won Lee

IBM T. J. Watson Research Center

Research was sponsored by the U.S. Army Research Laboratory and the U.K. Ministry of Defense and was accomplished under Agree
men
t Number

W911NF
-
06
-
3
-
0001. The views and conclusions contained in this document are those of the authors and should not be interpreted as

representing the

official policies, either expressed or implied, of the U.S. Army Research Laboratory, the U.S. Government, the U.K. Ministry
of
Defense or the U.K.

Government.

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
2

Invited talk at KAIST, 4/30/2007

Credits


Collaborators

V. Pappas, A. Tantawi, A. Beygelzemer (IBM)

S. Seshan (CMU)

P. Lio, J. Crowcroft (Cambridge)

M. Gerla (UCLA)

A. Swami (ARL), T. McCutcheon (DSTL)


Slide credits

A. Tantawi, V. Pappas (IBM)

U. Lee, M. Gerla (UCLA)


IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
3

Invited talk at KAIST, 4/30/2007

What is ITA?


I
nternational
T
echnology
A
lliance for Network and
Information Sciences



Large scale long
-
term research program supported by US
ARL and UK MOD

10 years, 24 institutions in US and UK



Four main technical areas (TAs)

network theory, security of a system of systems, sensor information
processing, and coalition planning


IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
4

Invited talk at KAIST, 4/30/2007

The ITA Vision


A US/UK Alliance conducting an open collaborative
research focused on network science by:


Creating an international collaborative research culture


Academia, Industry, Government in US and UK


Innovative multidisciplinary approaches


Developing ground
-
breaking fundamental sciences


Making an impact on coalition military effectiveness


Develop understanding the fundamentals of military networks


not
just computer networks, but also logical and social networks


Jointly address major research challenges


Networking & Security & Sensor Processing & Decision making

5

U.S.

Gov
.

Industry

Academia

U.K.

Gov.

INDUSTRY

9.
BBNT Solutions LLC

10.
The Boeing Corporation

11.
Honeywell Aerospace Electronic Systems

12.
IBM Research

13.
Klein Associates

ACADEMIA

1.
Carnegie Mellon University

2.
City University of New York

3.
Columbia University

4.
Pennsylvania State University

5.
Rensselaer Polytechnic Institute

6.
University of California Los Angeles

7.
University of Maryland

8.
University of Massachusetts

INDUSTRY

8.
IBM UK

9.
LogicalCMG

10.
Roke Manor Research Ltd.

11.
Systems Engineering


& Assessment Ltd.

ACADEMIA

1.
Cranfield University, Royal Military

College of Science, Shrivenham

2.
Imperial College, London

3.
Royal Holloway University of London

4.
University of Aberdeen

5.
University of Cambridge

6.
University of Southampton

7.
University of York

7

10

6

4

2

8

5

3

1

9

13

12

11

1

2

3

4

5

6

7

8

9

10

11

Team Overview

6

International Technology Alliance in
Network and Information Sciences


Collaborative Alliance Managers/Consortium Managers


Jay Gowens (ARL)



Jack Lemon (MoD)


Dinesh Verma (IBM)



Dave Watson (IBM
-
UK)

Security Across a
System
-
of
-
Systems


Trevor Benjamin (Dstl)

Greg Cirincione (ARL)

John McDermid (York U)

Dakshi Agrawal (IBM)

Network Theory





Ananthram Swami (ARL)

Tom McCutcheon (Dstl)

Don Towsley (U Mass)

Kang
-
Won Lee (IBM)

Sensor Information
Processing


Tien Pham (ARL)

Gavin Pearson (Dstl)

Thomas La Porta (PSU)

Vic Thomas (Honeywell)

Distributed Coalition
Planning



Jitu Patel (Dstl)

Mike Strub (ARL)

Nigel Shadbolt (SHamp)

Graham Bent (IBM)

7

International Technology Alliance in
Network and Information Sciences


Collaborative Alliance Managers/Consortium Managers


Jay Gowens (ARL)



Jack Lemon (MoD)


Dinesh Verma (IBM)



Dave Watson (IBM
-
UK)

Security Across a
System
-
of
-
Systems


Trevor Benjamin (Dstl)

Greg Cirincione (ARL)

John McDermid (York U)

Dakshi Agrawal (IBM)

Network Theory





Ananthram Swami (ARL)

Tom McCutcheon (Dstl)

Don Towsley (U Mass)

Kang
-
Won Lee (IBM)

Sensor Information
Processing


Tien Pham (ARL)

Gavin Pearson (Dstl)

Thomas La Porta (PSU)

Vic Thomas (Honeywell)

Distributed Coalition
Planning



Jitu Patel (Dstl)

Mike Strub (ARL)

Nigel Shadbolt (SHamp)

Graham Bent (IBM)


Policy Based Security
Management


Calo, IBM

Energy Efficient
Security
Architectures and
Infrastructures


Paterson, Royal
Holloway

Trust and Risk
Management in
Dynamic Coalition
Environments


McDermid, York


Theoretical
Foundations for
Analysis/Design of
Wireless and Sensor
Networks


Towsley, U Mass

Interoperability of
Wireless Networks
and Systems


Lee, IBM

Hancock, RMR

Biologically
-
Inspired
Self
-
Organization in
Networks


Lio, Cambridge

Pappas, IBM


Quality of Information
of Sensor Data


Bisdikian, IBM

Task
-
Oriented
Deployment of
Sensor Data
Infrastructures


La Porta, Penn State

Complexity
Management of
Sensor Data
Infrastructures


Szymanski, RPI


Mission Adaptive
Collaborations


Poltrock, Boeing

Command Process
Transformation and
Analysis


Sieck, Klein Assoc

Shared Situational
Awareness and the
Semantic Battlespace
Infosphere


Shadbolt, Southhampton

Wagget, IBM

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
8

Invited talk at KAIST, 4/30/2007



Network Theory (Towsley U. Mass, Lee IBM)


Fundamental underpinnings for adaptive
networking to support complex system
-
of
-
systems

P1
Theoretical foundations for design of wireless and
sensor networks (Towsley, U. Mass)

P2
Interoperability of wireless networks and systems
(Lee IBM
-
US/Hancock, RMR)

P3
Biologically
-
inspired self
-
organization in networks

(Lio Cambridge/Pappas IBM
-
US)

Strategies for delivering traffic in duty
-
cycling networks

Power reduction by cooperative
transmission

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
9

Invited talk at KAIST, 4/30/2007

Biologically
-
inspired Networking


Why do computer scientists (who work in wireless
networking) look for biological inspirations?


At
high level
, there is a parallel between the two, e.g.

Topology and spatial characteristics

Dynamics and mobility vs. ants or insects foraging

Data diffusion vs. disease spreading

Robust design vs. self healing systems

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
10

Invited talk at KAIST, 4/30/2007

Holy Grail


Develop
simple

algorithms that uses only local knowledge,
which result in desirable
global

properties


Some network algorithms are like that (not necessarily
biological)

TCP congestion control [Jacobson88]

Randomized duty cycling [Godfrey04]

Coloring
-
based resource allocation [Ko05]

Time synchronization of nodes [Degesys07]


IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
11

Invited talk at KAIST, 4/30/2007

However,


Wireless networks are not graphs



They are even different from conventional networks

Physical characteristics

Medium access (resource sharing)

Routing

Dynamics and mobility



There is no single kind of wireless networks

Cellular, MANET, wireless mesh, sensors, aquatic, etc.


IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
12

Invited talk at KAIST, 4/30/2007

Issues at Various Layers


Physical layer

Antenna technologies


directional, MIMO, cooperative

Power control


also an issue at MAC, network layers


MAC layer in wireless

Hidden terminal problem, exposed terminal problem

Fairness in MAC


Network layer

Myriads of ad hoc routing protocols


Proactive, reactive, geographical, hierarchical, hybrid

Multicasting and broadcasting issues

Store
-
and
-
forward

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
13

Invited talk at KAIST, 4/30/2007

Bio
-
inspiration is Not Bio
-
emulation

X

O

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
14

Invited talk at KAIST, 4/30/2007

Topic of Today: Two Ongoing Research Activities


MANET/sensor net topology control

IBM / CMU


Urban sensing and data diffusion

IBM / UCLA / Cambridge

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
15

Invited talk at KAIST, 4/30/2007

MANET Topology Control


Problem Definition

How to configure low
-
level device parameters in order to achieve a
network structure with a set of desirable characteristics?


Characteristics

Connectivity

Network capacity

Energy consumption

Path latency

Robustness/Resilience

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
16

Invited talk at KAIST, 4/30/2007

Configurable Parameters


Transmission power


Carrier sense threshold

MAC, packet transmission


Radio channel allocation

Multiple channel / multiple interface


Antenna characteristics:

Multiple
-
input multiple
-
output (MIMO)

Directional, onmi
-
directional


IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
17

Invited talk at KAIST, 4/30/2007

Example: Transmission Power

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
18

Invited talk at KAIST, 4/30/2007

Topology Properties

Pictures from CBTC paper (ToN 05)

Densely Connected

Sparsely Connected

Small Hop
-
Count

High Power Consumption

Robust to Node Failures

Transmission Interference

Large Hop
-
Count

Low Power Consumption

Prone to Node Failures

Low Interference

What makes a

good topology ?

Small Hop
-
Count

Low Power Consumption

Robust to Node Failures

Low Interference

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
19

Invited talk at KAIST, 4/30/2007

Good Topology: Application
-
Driven


Application
-
driven topology control:

Application
-
specific metrics

Placement of services

Compatibility matrix

Dense

Sparse

Real
-
Time

Messaging

Dense

Sparse

Long

Short

Application
-
Type

Traffic
-
Matrix

Mission
-
Duration

Topology

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
20

Invited talk at KAIST, 4/30/2007

How to Build Good Topologies (1)


Can we get insights from wired networks?


How about biological insights


neural networks,
galleries of insect colony?

[Li04]

Preferential Attachment

HOT Model

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
21

Invited talk at KAIST, 4/30/2007

How to Build Good Topologies (2)


What do we need to consider in MANET topology?

MANET: spatial constrains

Technological Constrains: shared medium, energy consumption

Node Mobility

Time Scale



As a result:

Internet: scale
-
free

MANET: RGG? Clustered?

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
22

Invited talk at KAIST, 4/30/2007

How to Build Good Topologies (3)


Current Approaches:

Minimize transmission power


NP
-
hard problem (for 2D and up)

Minimize interference with channel allocation


NP
-
hard problem

Minimize energy stretch of a path


Relative Neighbor Graphs, Gabriel Graph, Yao Graph



IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
23

Invited talk at KAIST, 4/30/2007

Duty Cycling in Wireless Networks


Power saving


longevity of mission lifetime


Impacts the performance

Sensor coverage

Connectivity

Routing delay


Mathematical modeling to provide insights for management

How to control the fraction of active nodes

Localized duty cycling decision


predictable global behavior

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
24

Invited talk at KAIST, 4/30/2007

Related Work


SPAN (Chen01)

Makes local randomized decision to join a forwarding backbone based on
the estimate how much it will benefit the neighbors


GAF (Xu01)

Sets up a virtual grid based on location information, and only one node in a
grid becomes active


STEM (Schurgers02)

Nodes awaken sleeping neighbors when they need to forward data using
beacons on a dedicated signaling channel


NAPS

Local randomized algorithm based on number of neighbors with an aim to
achieve global connectivity

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
25

Invited talk at KAIST, 4/30/2007

Modeling Duty Cycling Networks


Consider two states: active, sleeping


Each node makes local decision based on:

Its own probability to become active

States of immediate neighbors: pulling or pushing


We are interested in the steady state

Model as a spatial process


IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
26

Invited talk at KAIST, 4/30/2007

Modeling Duty Cycling Networks


Spatial Process

n

=

(

n

1

;

n

2

;

¢

¢

¢

;

n

J

)

J

s

i

t

e

s

n

j

a

t

t

r

i

b

u

t

e

(

s

t

a

t

e

)

o

f

s

i

t

e

j

2

f

1

;

2

;

¢

¢

¢

;

J

g

,

n

j

2

N

j

S

s

t

a

t

e

s

p

a

c

e

,

S

=

N

1

£

N

2

£

¢

¢

¢

£

N

J

¼

p

r

o

b

a

b

i

l

i

t

y

d

i

s

t

r

i

b

u

t

i

o

n

¼

:

S

!

(

0

;

1

)

a

n

d

P

n

2

S

¼

(

n

)

=

1

n

i

s

c

a

l

l

e

d

a

r

a

n

d

o

m

¯

e

l

d

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
27

Invited talk at KAIST, 4/30/2007

Connectivity Model

G

c

o

n

n

e

c

t

i

v

i

t

y

g

r

a

p

h

o

f

s

i

t

e

s

(

J

;

E

)

G

¡

j

s

e

t

o

f

s

i

t

e

s

i

n

G

o

t

h

e

r

t

h

a

n

j

g

j

s

e

t

o

f

n

e

i

g

h

b

o

r

s

o

f

s

i

t

e

j

n

i

s

a

M

a

r

k

o

v

¯

e

l

d

i

f

P

(

n

j

j

n

G

¡

j

)

=

P

(

n

j

j

n

g

j

)

,

j

2

f

1

;

2

;

¢

¢

¢

;

J

g

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
28

Invited talk at KAIST, 4/30/2007

Probability Distribution


Product Form

F

o

r

M

a

r

k

o

v

¯

e

l

d

n

,

¼

h

a

s

t

h

e

p

r

o

d

u

c

t

f

o

r

m

¼

(

n

)

=

B

¦

C

2

C

µ

C

(

n

C

)

;

n

2

C

w

h

e

r

e

C

µ

G

i

s

a

s

i

m

p

l

e

x

(

a

s

e

t

o

f

f

u

l

l

y

c

o

n

n

e

c

t

e

d

e

d

g

e

s

)

C

s

e

t

o

f

s

i

m

p

l

i

c

e

s

o

f

G

µ

C

(

n

C

)

a

f

u

n

c

t

i

o

n

o

f

n

C

,

C

2

C

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
29

Invited talk at KAIST, 4/30/2007

Geometric Reversible Spatial Process

S

u

p

p

o

s

e

t

h

a

t

N

j

=

f

1

;

2

;

¢

¢

¢

;

N

g

q

(

n

;

T

m

j

n

)

=

¸

(

n

j

;

m

)

Á

(

n

j

)

r

Ã

(

m

)

r

0

w

h

e

r

e

T

m

j

n

=

(

n

1

;

n

2

;

¢

¢

¢

;

n

j

¡

1

;

m

;

n

j

+

1

;

¢

¢

¢

;

n

J

)

a

n

o

p

e

r

a

t

o

r

w

h

i

c

h

c

h

a

n

g

e

s

t

h

e

a

t

-

t

r

i

b

u

t

e

o

f

s

i

t

e

j

t

o

m

¸

(

n

j

;

m

)

i

n

t

r

i

n

s

i

c

t

e

n

d

e

n

c

y

o

f

a

s

i

t

e

t

o

c

h

a

n

g

e

f

r

o

m

n

j

t

o

m

r

(

r

0

)

n

u

m

b

e

r

o

f

s

i

t

e

s

n

e

i

g

h

b

o

r

i

n

g

j

w

i

t

h

a

t

t

r

i

b

u

t

e

s

n

j

(

m

)

Á

(

n

j

)

(

Ã

(

m

)

)

e

x

t

r

i

n

s

i

c

t

e

n

d

e

n

c

y

o

f

a

s

i

t

e

t

o

c

h

a

n

g

e

f

r

o

m

n

j

(

t

o

m

)

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
30

Invited talk at KAIST, 4/30/2007

Geometric Reversible Spatial Process

T

h

e

e

q

u

i

l

i

b

r

i

u

m

d

i

s

t

r

i

b

u

t

i

o

n

i

s

¼

(

n

)

=

B

¦

N

n

=

1

®

(

n

)

M

(

n

)



Ã

(

n

)

Á

(

n

)

¸

R

(

n

)

w

h

e

r

e

M

(

n

)

n

u

m

b

e

r

o

f

s

i

t

e

s

w

i

t

h

a

t

t

r

i

b

u

t

e

n

R

(

n

)

n

u

m

b

e

r

o

f

e

d

g

e

s

w

i

t

h

b

o

t

h

e

n

d

s

i

t

e

s

h

a

v

i

g

a

t

t

r

i

b

u

t

e

n

a

n

d

®

(

n

)

i

s

t

h

e

n

o

n

z

e

r

o

s

o

l

u

t

i

o

n

t

o

®

(

n

)

¸

(

n

;

m

)

=

®

(

m

)

¸

(

m

;

n

)

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
31

Invited talk at KAIST, 4/30/2007

Example: Simple Duty Cycling

N

=

f

0

;

1

g

¸

(

1

;

0

)

=

¸

¸

(

0

;

1

)

=

¹

®

=

¹

=

¸

Á

(

0

)

=

Á

(

1

)

=

1

Ã

(

0

)

=

¡

Ã

(

1

)

=

¢

T

h

e

n

,

¼

(

n

)

=

B

®

M

(

1

)

¡

R

(

0

)

¢

R

(

1

)

T

h

r

e

e

p

a

r

a

m

e

t

e

r

s

:

®

,

¡

,

a

n

d

¢

// 0: sleeping, 1: active

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
32

Invited talk at KAIST, 4/30/2007

Analytical Result


1k Node, 10k Edge RG

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
33

Invited talk at KAIST, 4/30/2007

Impact of Self


Increasing
α

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
34

Invited talk at KAIST, 4/30/2007

Impact of Self


Full Spectrum

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
35

Invited talk at KAIST, 4/30/2007

Impact of
Ψ
(0)



Increasing
γ

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
36

Invited talk at KAIST, 4/30/2007

Impact of
Ψ
(0)



Full Spectrum

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
37

Invited talk at KAIST, 4/30/2007

Impact of
Ψ
(1)



Full Spectrum

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
38

Invited talk at KAIST, 4/30/2007

Impact of Both
Ψ
(0)

and
Ψ
(1)


IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
39

Invited talk at KAIST, 4/30/2007

Controlling Node Activities

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
40

Invited talk at KAIST, 4/30/2007

Connectivity Graphs

C

G

F

E

D

B

A

H

sample graph

linear graph

3

5

4

2

1

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
41

Invited talk at KAIST, 4/30/2007

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
42

Invited talk at KAIST, 4/30/2007

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
43

Invited talk at KAIST, 4/30/2007

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
44

Invited talk at KAIST, 4/30/2007

Two Ongoing Research Activities


MANET/sensor net topology control

IBM / CMU


Urban sensing and data diffusion

IBM / UCLA / Cambridge

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
45

Invited talk at KAIST, 4/30/2007

Epidemic Style Data Diffusion in Vehicular
Sensor Networks (VSNs)

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
46

Invited talk at KAIST, 4/30/2007

Vehicular Sensor Applications




Smart
-
mob
-
approach for proactive urban monitoring using
VSN

Smart mobs: people with shared interests and goals persuasively
and seamlessly cooperate using wireless mobile devices



Environment

Traffic congestion monitoring

Urban pollution monitoring



Civic and Homeland security

Forensic data for accidents or crime sites

Terrorist alerts


IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
47

Invited talk at KAIST, 4/30/2007

Accident Scenario: Storage and Retrieval


Designated cars:


Continuously collect images on the street (store data locally)

Process the data and detect

an event

Classify the event as Meta
-
data

(Type, Option, Location, Vehicle ID)

Post it on distributed index


Police (agents) retrieve data from designated cars

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
48

Invited talk at KAIST, 4/30/2007

How to Retrieve the Data?


Upload to nearest AP (Cartel project, MIT)


Epidemic diffusion (our approach)

Mobile nodes

periodically broadcast
meta
-
data
of events to their
neighbors

A
mobile

agent

(e.g. the police) queries nodes and harvests events

Data dropped when stale and/or geographically irrelevant


IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
49

Invited talk at KAIST, 4/30/2007

General Problem


Three phases of urban sensing & harvesting

Meta
-
data Dissemination

Meta
-
data Harvesting

Data Access



Bio inspirations:

Pheromone trails (ants foraging)

Chemotaxes (bacterial foraging)


Motion patterns (called taxes) that the bacteria generates in prescreens
of chemical attractants and repellants (nutrition gradient) (e.g., E. Coli)


IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
50

Invited talk at KAIST, 4/30/2007

Communications of Ants


Decoupling of foraging and recruitment

Pheromone trail: route to food

Dance and physical contract: recruitment of additional foragers


Types of pheromone trails

Non
-
volatile, volatile, short
-
lived repellent


Sound, physical contacts (time
-
space constraints)

Antenna, vibration, displays, dances, waggling, jerking



Pharaoh’s ants,
Monomorium
pharaonis,

form branching
networks of pheromone trails.
There the network has been
formed on a smoked glass surface
to aid visualization


(Image courtesy of Duncan
Jackson)

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
51

Invited talk at KAIST, 4/30/2007

Meta
-
data Dissemination


Meta
-
data creation

Format: (Location and timestamp, data type, variable size info)

Optional local processing, e.g. Recognizing license plates + vehicle type



Dissemination

Periodically broadcast to neighbors

Can be encrypted for security/privacy issues



Prioritization

Temporal, spatial

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
52

Invited talk at KAIST, 4/30/2007

Meta
-
data Harvesting


Gradient
-
based foraging

Vehicle density in urban grids is
non
-
uniform


More vehicles, more information: Agents are attracted via this
info gradient

Need to avoid local maxima



Reinforcement learning

Learn the mobility patterns over time

Data
-
mining results can provide “feedback” to the foraging algorithm



Multiple agents

Harvesting area should be divided to minimize interference

For example, based on contact history (as repellents)


IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
53

Invited talk at KAIST, 4/30/2007

Data Access


Collection by agents

Similar to LER with actual mobility

Factors: physical speed of agents; coordinated swarming of agents



Collection by networks

Multi
-
hop pulling via Last Encounter Routing (LER)


IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
54

Invited talk at KAIST, 4/30/2007

Evaluation


Simulation Setup

NS
-
2 simulator

802.11: 11Mbps, 250m tx range

Average speed: 10 m/s

Mobility Models


Random waypoint (RWP)


Real
-
track model (RT) :


Group mobility model


merge and split at intersections



Westwood

map

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
55

Invited talk at KAIST, 4/30/2007

Meta
-
data Harvesting Delay with RWP


Higher mobility decreases harvesting delay

Time (seconds)

# of Harvested Summaries

V=25m/s

V=5m/s

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
56

Invited talk at KAIST, 4/30/2007

Harvesting Results with Real Track


Restricted mobility results in larger delay

Time (seconds)

# of Harvested Summaries

V=25m/s

V=5m/s

IBM T J Watson Research Center

© 2007 IBM Corporation

Slide
57

Invited talk at KAIST, 4/30/2007

To sum up


ITA opportunity

International collaborative research on interesting topics



More understanding is required

Biology/physics camp vs. computer networks


BIOWIRE workshop (Cambridge, UK)


Network
-
based modeling, simulation vs. Analysis based on ODE

Mobility model



Questions?