coursWN1 Lucile2014

Course: Content Distribution in
Wireless Networks
Topic: Network Coding and Opportunistic
routing in mobile social networks
Master 2 IFI Ubinet 2014-2015
L. Sassatelli
[email protected]
1
Ubiquitous / Pervasive / Mobile

computing
In 1991, Marc Weiser (Xerox PARC), formulated his vision of a new area in
computer science: ubiquitous computing.
To enable people to move around and interact with computers more
naturally than they were used to do. Computers should become good,
invisible tools. In his sense, an invisible tool is one that does not draw the
users attention towards itself.
Another term: pervasive computing.

Embedding computation power into the
environment.
Pervasive computing does not take node or
user mobility into account.
A. Heinemann, Collaboration in Opportunistic Networks, PhD dissertation, Darmstadt 2007
Opportunistic networks
Increasing integration of wireless short-range
communication technologies (Bluetooth, 802.11 WiFi) into
mobile devices
-> spontaneous communication, interaction and
collaboration are possible.
Spontaneous communication: opportunistic networking
Promising evolution in mobile ad-hoc networking.
Formed by mobile devices communicating while users are
in close proximity.
There are two prominent characteristics present in
opportunistic networks: 1) A user provides his personal
device as a network node. 2) Users are a priori unknown to
each other.
About
This course is about:
Network coding as a new networking paradigm
Fundamentals
Applications of NC to wireless networks, to
opportunistic networking
Opportunistic routing in MANET

Mobility models: social networks in motion
Opportunistic routing in heterogeneous MANETs:
strategies and performance analysis
Not about
This course is not about:
Software implementations
Security and privacy issues of MANETs
Incentives to encourage collaboration,
competition in mobile networks (game theory)
Outline
I.
Network coding: beyond routing

1.
2.
II.
Introduction
Application of NC to wireless networks: Xor in the air
Opportunistic routing
1.
2.
Motivation for opportunistic networking

Routing in DTN with uniform mobility
1.
2.
3.
3.
Routing in heterogeneous DTN: mobile social networks

1.
2.
3.
4.
Some uniform mobility models and their characterization

Two-hop routing, epidemic routing and performance modeling
Spray-and-Wait routing: trade-off between resource consumption and performance
Introduction to mobile social networks: the small-world effect, characterization of human
mobility
Utility-based routing (UBR): accounting for social relationships
On what UBR builds on: the problem of mapping contacts to graph, community detection
and decentralized decisions
Network coding under Spray-and-Wait in DTN
Network coding
Network coding: generalization of routing the intermediate nodes can modify the
payload of packets they have to transfer
Transport networks
source s
Oriented valued graph G = (V, E, c)

c(p, q): capacity of edge (p, q)
f(p, q): rate or flow of edge (p, q)
Examples:
Water pipes
Pipelines
Transportation lanes
Freight traffic
Communication networks
Sink t
Conditions
Capacity c : V x V R with c(p, q) 0 and if (p, q) not in E then c(p, q) = 0
Flow f : V x V R
Source s V, sink (terminal) t V
Accessiblity: all the vertices are on a path from s to t
Constraint of capacity:
for all p, q V, f (p, q) c (p, q)
c(p, q)
f (p, q)
Anti-symmetry:
for all p, q V, f (q, p) = - f (p, q)
Flow conservation:
for all p V\{s,t}, (f (p, q) | q V ) = 0
p
7
Flow
source s
Value of the flow

| f | = ( f (p, t) | p V )
that arrives to the sink
Properties
for all p V, f (p, p) = 0
for all q V, ( f (p, q) | p V ) = 0
flow / capacity
sink t
q
10
Problem of max flow

Given an oriented valued graph: G = (V, E, c)
Compute a maximum flow allocation:
set f(p,q), for all p,q E, such that |f| is maximum
source s
capacities
A
6
B 3
3 C
A
6/6
5/8
B 0/3 2/3 C
3
A maximum flow?
| f | = 11
6/6
3/3
8/8
3/6
sink t
11
Cut
s
(X, Y) is a cut of G iif:

(X, Y) partition of V with s X, t Y
A
6/6
Capacity of the cut:

c(X, Y) = ( c(x, y) | x X, y Y )
Flow of the cut
f (X, Y) = ( f (x, y) | x X, y Y )
6/8
B 1/3 3/3 C
5/6
3/3
8/8
4/6
F
t
| f | = 12
X = {A,B,C,D} Y = {E,F} c(X,Y) = 14 f (X,Y) = 12
12
Properties
Properties:
1. For any cut (X,Y), we have f (X,Y) = | f |
2. For any cut (X,Y), f (X,Y) c(X,Y)
3. f is a maximum flow (allocation) iif
there exists a cut (X0,Y0) such that |f|= c (X0,Y0)
->Theorem of min-cut max-flow:

The maximum flow is the minimum cut capacity.
13
Min-cut
X0 = {A,C}
Y0= {B,D,E,F}
c(X0,Y0) = 12
(X0,Y0) of min capacity
f (X0,Y0) = 12
Max flow
A
6/6
6/8
B 1/3 3/3 C
5/6
X0
3/3
8/8
4/6
Y0
t
| f | = 12
14
Multicast Problem
Butterfly Networks: Each
edges capacity is 1.
Max-Flow from A to D = 2
Max-Flow from A to E = 2
Multicast Max-Flow from A
to D and E = 1.5
Max-Flow for each
individual connection is not
achieved.
C
F
E
15
by A. Limmanee
Network Coding
Introduction
Linear Network Coding
Transfer Matrix
Network Coding Solution
Connection between an Algebraic Quantity
and a Graph Theoretic Tool
Finding Network Coding Solution
16
by A. Limmanee
Introduction: multicast with routing

b1 b2
A
b1
B
b1
b2
C
b2
F
b1
b1
b1
b2
D
b1 b2
b2
b2
b1
b2
E
b1 b2
17
by A. Limmanee
Introduction: multicast with network

coding
Ahlswede et al. (2000)
b1 b2
With network coding,

every sink obtains the
maximum flow.
Li et al. (2003)
Linear network coding is
enough to achieve the
maximum flow, for
multicast in directed
networks
b1
b2
b2
F
b1
b2
b 1 + b2
b1 + b2
D
b1 b2
R. Ahlswede, N. Cai, S. R. Li, and R. W. Yeung, Network information ow, IEEE
Transactions on Information Theory, vol. 46, no. 4, July 2000.
b1
b1 + b2
E
b1 b2
18
by A. Limmanee
Linear Network Coding

Random Processes in a Linear Network
Source Input: X (v, l ) x0 (v, l ), x1 (v, l ),...
Y (e) The
y0index
(e),is ya 1 (e),...
Info. along Edges:
Combination of
index
Combination
SinkWeighted
Output:
Z (ofv, l ) z0 (vtime
, l ),Weighted
z
(
v
,
l ),...adjacent
1
processes
from
processes generated at v
edges of e
Relationship between them
(v)
Y (e) l ,e X (v, l )
e comes out
of v
l 1
Z (v, j )
Weighted Combination
from all incoming
edges
Y (e)
e, e
e:head ( e) tail ( e )
e, j
e:head ( e) v
Y (e)
19
by A. Limmanee
Transfer Matrix
Let x ( X (v1 ,1), X (v1 ,2), X (v1 ,3)) Y (e1 ) 1,e1 X (v,1) 2,e1 X (v,2) 3,e1 X (v,3)
z (Z (v4 ,1), Ze1(v4 ,2), Z (v4 ,3)) Y (e2 ) 1,e2 X (v,1) 2,e2 X (v,2) 3,e2 X (v,3)
X (v1 ,1)
z x e M
v2
e5
Y (e3 ) 1,e3 X (v,1) 2,e3 X (v,2) 3,e3 X (v,3)
Z (v4 ,1) Y (e ) Y (e ) Y (e )
4
e1 ,e4
1
e2 , e 4
2
e1v,e44 e4 ,e7 Z (v4 ,2)
v1e1 ,e5 ee1 ,4e4 e4e,e66

Y (e5 ) e1 ,e5 Y (e1 ) e2 ,e5 Y (e2 )
X (vM
) A e ,e e ,e e ,e e ,e e ,eZ(vB4 ,3)
1 ,3
2 e
5
2 4
4 6
2 4
4 7
Y (e6 ) e3 ,e6 Y (e3 ) e4 ,e6 Y (e4 )
3
e7
v
e33 ,e6
e3 ,e7
Y (e7 ) e3 ,e7 Y (e3 ) e4 ,e7 Y (e4 )
0
1,e ,1 1,ee , 2 1,ee ,3
Z (v4 ,1) e5 ,1Y (e5 ) e6 ,1Y (e6 ) e7 ,1Y (e7 )
5
5
5
Z (v4 ,2) e5 , 2Y (e5 ) e6 , 2Y (e6 ) e7 , 2Y (e7 )

AB2,e 2,e 2,e
e6 ,1
e6 , 2
e6 , 3
Z (v4 ,3) e5 ,3Y (e5 ) e6 ,3Y (e6 ) e7 ,3Y (e7 )
3,ee ,1 3,ee , 2 3,ee ,3

20
by A. Limmanee
7
7
7
X (v1 ,2)
Network Coding Solution

z xM
We want z x
e ,e e ,e e ,e e ,e e ,e
Choose A to be an
M A e ,e e ,e e ,e e ,e e ,e B
identity matrix.
e ,e
e ,e
0
Choose B to be the
1,e 1,e 1,e
inverse of
NETWORK CODING
A 2,e 2,e 2,e ,

IF
SOLUTION EXISTS
3, e
3, e
3, e
DETERMINANT OF M IS
e ,1 e , 2 e ,3
e ,e
e ,e
NON-ZERO
1 5
1 4
4 6
2 5
2 4
4 6
3 6
1 4
2 4
4 7
4 7
3 7
1 5
1 4
4 6
2 5
2 4
4 6
B e6 ,1 e6 , 2 e6 ,3
e7 , 3
e7 ,1 e7 , 2
3 6
1 4
2 4
4 7
4 7
3 7
by A. Limmanee
21
Connection between an Algebraic

Quantity and a Graph Theoretic Tool
Let a linear network be given with source node ,
sink node , and a desired connection c , , ,
of rate R(c). The following three statements are
equivalent.
1. The connection c , , , is possible.
2. The Min-Cut Max-Flow bound is satisfied
3. The determinant of the R(c) R(c) transfer matrix M
is non-zero over the ring F2*,l,e,,e,e,,e,j,+.
Ralf Koetter and Muriel Mdard. An algebraic approach to network coding. IEEE/ACM
Trans. on Netw. 11, 5, October 2003.
22
by A. Limmanee
Finding Network Coding Solution

Koetter and Medard (2003): Greedy Algorithm
Let a delay-free communication network G and a
solvable multicast problem be given with one source
and N receivers. Let R be the rate at which the source
generates information. There exists a solution to the
network coding problem in a finite field F2 with
m
m log 2 ( NR 1)
Ralf Koetter and Muriel Mdard. An algebraic approach to network coding. IEEE/ACM
Trans. on Netw. 11, 5, October 2003.
23
by A. Limmanee
Random Network Coding

is the number of edges and d is the number of sink nodes
--> Choosing the coding coefficient uniformly at random in , with q large enough, is
sufficient to ensure high probability of decoding at the sink(s)
Or: For a fixed success probability, the field size needs to be on the order of the number of
links multiplied by the number of receivers d.
--> Fully decentralized
RLNC achieves robustness to link failures
T. Ho, R. Koetter, M. Mdard, D. R. Karger, and M. Effros, The benets of coding over routing in a randomized
setting, in IEEE International Symposium on Information Theory, 2003.
T. Ho and D.S. Lun, Network Coding: An Introduction, Cambridge University Press, 2008
24
On lossy channels: NC for unicast

improves rate AND delay
To cope with erasures on a channel with , we need to send N=K+M pkts

to retrieve K info pkts (M such that (K+M)(1- )=K).
The additional M pkts are not replications, but function of all K info pkts.
The K info pkts must be retrieved from the NN received pkts. Then there
are 3 solutions for the 2 links problem :
25
On lossy channels: NC for unicast

improves rate AND delay
S sends N pkts such that N(1- 12)(1- 23)=K. Then the end-toend rate is K/N=(1- 12)(1- 23)
If the relay is able to decode (retrieve all K info pkts from the
N1 received) then
S can send N1 such that N1(1- 12)=K
R can send N2 such that N2(1- 23)=K
Then the E2 rate is min(1- 12,1- 23)N(1- 12)(1- 23)
BUT: such block coding entails delay
RLNC at relay: R generates pkts on the fly by mixing those

received from S, w/o decoding
--> no delay and max rate
26
Inter-session network coding

Intra-session network coding
Inter-session network coding
a+b
The COPE example
improves the throughput of

lossless multicast sessions,
and of lossy unicast or
multicast sessions
The butterfly example
is necessary to achieve
optimal throughput in general
27
Practical Issues
Network Delay
Centralized Knowledge of Graph Topology
Packet Loss
Link Failures
Change in Topology or Capacity
28
by A. Limmanee
Outline
I.

1.
2.
II.
Introduction
1.
2.

1.
2.
3.
3.

1.
2.
3.
4.

mobility

29
XORs in The Air: Practical Wireless

Network Coding
S. Katti, H. Rahul, W. Hu, D. Katabi, M. Mdard and J. Crowcroft. XORs in the air:
Practical wireless network coding. In Proceedings of SIGCOMM 2006.
30
The problem
Wireless networks are highly resource
constrained
Bandwidth is the most expensive
Power is sometimes an issue too
--> Serious problems for mesh networks
How to optimize throughput?

Can we send more information?
Can we reduce bandwidth requirement?
--> Do both at the same time?
31
by Wenjun Hu
An information exchange scenario

Relay
Alice
Alices packet
Bobs packet
Bob
Bobs packet
Alices packet
Multi-hop unicast requires 4 transmissions

Can we do better?
32
by Wenjun Hu
Can Network Coding help? - An idea

XOR
Relay
Alice
Alices packet
Bobs packet
Bob
Bobs packet
Alices packet
3 transmissions instead of 4
Saves bandwidth & power
33% throughput increase
by Wenjun Hu
33
The COPE approach

Considers multiple unicast flows
Generalizes the duplex flow scenario
Opportunistic coding using local info

Overhear packets to increase coding gain
Online, distributed and deployable
Emulation and testbed results

First real-world implementation
34
by Wenjun Hu
COPE: Opportunistic Coding Protocol

Alice
Bob
Bob
Charlie
Charlie
Alice
XOR
Alice
Alices packet
Bobs packet
Charlies packet
Charlie
Charlies packet
Alices packet
Bobs packet
XOR
Relay
Bob
Bobs packet
Charlies packet
Alices packet
35
by Wenjun Hu
How it works(Cont.)
Relay Encoding
Checks packets in queue
Combines packets traversing the same three hops in

opposite directions
Metadata in a header between MAC and IP
Broadcast encoded packets
Alice/Bob Decoding
Keep copies of sent packets
Detect the extra header (decoding info)
Retrieve the right packet to decode
Distributed and local action only!

36
by Wenjun Hu
Generalize to COPE
Nodes snoop on the medium
Reception reports to neighbours
When encoding
Identify what packets neighbours have
Reception reports and guesses
Encode as many packets as possible

Provided intended recipients can decode them
Still distributed and local action only!

37
by Wenjun Hu
The importance of being opportunistic

Opportunistic coding
Only encode if packets in queue
No delay penalty
Insensitive to flow characteristics
Opportunistic listening
Helps create more coding opportunities
38
by Wenjun Hu
Pseudo-broadcast
COPE gain is from broadcast medium
But 802.11 broadcast doesnt work!
No reliability scheme to mask collision loss
Send packets at lowest bit rate
May actually reduce throughput!
Pseudo-broadcast
Send encoded packets as if unicast
Other neighbours overhear
Benefit as a unicast packet
39
by Wenjun Hu
Implementation
A shim between MAC and IP
Agnostic to protocols above/below
Emulations
General COPE
Emsim (part of Emstar) environment
Testbed
Based on the Alice/Bob scenario
Extension to Roofnet code (in Click)
40
by Wenjun Hu
Emulation Scenario
100 nodes in 800m x 800m
Consider range ~50m
Random senders/receivers
Senders always backlogged
Bit rate at 11 Mb/s
Geographic routing
Metric: end-to-end data traffic throughput
over all flows
41
by Wenjun Hu
Emulation performance
Throughput (KB/s)
Coding always outperforms no-coding

42
by Wenjun Hu
Testbed setup
Indoor PCs with 802.11b cards
Intersil Prism 2.5 802.11b chipset
Connected to omni-directional antenna
RTS/CTS disabled
802.11 ad hoc mode
Randomly chosen 3 nodes from testbed

Static routes
End nodes send UDP traffic to each other
43
by Wenjun Hu
Testbed results
Ratio of Throughput with Coding to No-Coding
2
1.8
1.6
1.4
1.2
1
1
5.5
11
802.11 Bit Rates (Mb/s)
Encoding almost doubles the throughput

44
by Wenjun Hu
Why more than 33%?

Relay
Alice
Bob
MAC is fair -> 1/3 BW for each node

Without coding, relay needs twice as much
bandwidth as Alice or Bob
With coding, all nodes need equal bandwidth
45
by Wenjun Hu
Summary
Opportunistic approach allows practical
integration of network coding into current
stack
Throughput can double in practice
Cross-layer effects
Congestion plays in our favour
First implementation of network coding in a

wireless environment
46
by Wenjun Hu
Outline
I.

1.
2.
II.
Introduction
1.
2.

1.
2.
3.
3.

1.
2.
3.
4.

mobility

47
Opportunistic networks
Increasing integration of wireless short-range
communication technologies (Bluetooth, 802.11 WiFi) into
mobile devices
-> spontaneous communication, interaction and
collaboration are possible.
Spontaneous communication: opportunistic networking
Promising evolution in mobile ad-hoc networking.
Formed by mobile devices communicating while users are
in close proximity.
There are two prominent characteristics present in
opportunistic networks: 1) A user provides his personal
device as a network node. 2) Users are a priori unknown to
each other.
48
Disruption/Delay Tolerant Mobile Ad

Hoc Networks: Overview
Goal of DT-MANETs: allowing communication between
mobile users, even in the absence of infrastructure, and
when a contemporaneous path from S to D does not exist
necessarily (due to disconnections and/or high mobility).
Such networks can be
Interference-limited
Or connectivity-limited
--> use of nodes as relays to achieve end-to-end

communication: Store-Carry-And-Forward paradigm
entails a certain communication delay
--> this MANETs are also referred to as Delay Tolerant
Networks (DTNs).
49
Pocket-switched networks
Thank you but you are in
the opposite direction!
I have 100M bytes of

data, who can carry for
me?
I can also carry for

you!
Give it to me, I have 1G

bytes phone flash.
Dont give to me! I

am running out of
storage.
Reach an access
point.
There is one in
my pocket
Search La
Bonheme.mp3 for me
Internet
Finally, it
arrives
Search La
Bonheme.mp3 for me
Search La
Bonheme.mp3 for me
J. Crowcroft,P. Hui, A. Chaintreau, J. Scott, R. Gass, C. Diot, Pocket Switched Networks, Univ. of Cambridge & Intel Research
50
Applications
Provide network access to remote communities
Provide cheaper content access by file exchange in ad hoc mode
Offload the telcos networks
Growing demand for contents --> infrastructure-centric networking paradigm has
limitations
--> promising alternative
to offload the telcos networks: exploit the

user interactions to convey information: Ad hoc Networks
51
Limitation of cellular networks
http://www.nytimes.com/2009/09/03/technology/companies/03att.html
http://venturebeat.com/2009/05/11/iphone-users-eating-up-atts-network/
52
DTN : Real-world applications

Ring Road: constellation of nanosatellites to provide low-cost access,
bypass geographic and/or political obstacles on earth
Bytewalla: connecting african villages, Android phones
Saami Network Connectivity: provides Internet access to a nomadic
community
Liberouter: combines the SCAMPI platform with a Raspberry Pi to
produce a cheap opportunistic router - also on Google Play
DTN-Bone: effort of the DTNRG to establish a worldwide collection of
nodes running DTN bundle agents and applications
https://www.youtube.com/playlist?list=PL4buVHalBRoPhqHKMBOU2yqPLtaN5vfKC
53
Outline
I.

1.
2.
II.
Introduction
1.
2.

1.
2.
3.
3.

1.
2.
3.
4.

mobility

54
Random mobility models:

Random walker
Equal probability to jump
in any direction
55
Random mobility models:

Random waypoint and Random direction
Node:
Picks next waypoint Xn+1 uniformly in area

Picks speed Vn uniformly in [vmin,vmax]
Moves to Xn+1 with speed Vn
Variant: Random direction
- Directions are uniformly

distributed (0, 2)
- Speeds (Vi) are uniformly
distributed (Vmin,Vmax)
Xn+1
- Travel times (Ti) are

exponentially /generally
distributed
Xn
56
J.-Y. Le Boudec, M. Vojnovid, Random Trip Mobility Models, Tutorial ACM Mobicom 2006
Inter-meeting times under random mobility

Inter-meeting times mobile/mobile have shown to follow
an exponential distribution
[Groenevelt et al.: The message delay in mobile ad hoc networks. Performance
Evaluation, 2005]
Pr{ X = x } = exp(- x)
CDF: Pr, X x - = 1 - exp(- x) ,
CCDF: Pr{ X > x } = exp(- x)
M. Ibrahim, Routing and Performance Evaluation of Disruption Tolerant Networks, PhD defense, UNS 2008
57
Outline
I.

1.
2.
II.
Introduction
1.
2.

1.
2.
3.
3.

1.
2.
3.
4.

mobility

58
Network model
Settings:
only a single unicast session
= 1/ : average number of meetings a node has per time unit
sparse DTN: =N remains constant as N increases
--> the network is connectivity-limited
mobility model: fast and uniform (RW, RWP,)
--> routing strategies must permit timely delivery of information to a certain

destination with high probability: use of replication
replication leads to energy and memory consumption

finite duration of radio contacts --> file may be split into packets
Objective: optimize the file transfer from S to D by minimizing both its delay, the
memory and energy required by the store, carry and forward process
59
Two-hop routing
Only the source can disseminate copies of the packet, the
relays can only deliver their
copy to the destination
D
D
S
D
D
D
A relay can deliver

only to the
destination
C
B
60
Epidemic routing
The packet is flooded into the network
D
D
D
E
D
D
S
D
D
D
C
B
Forwarding is not limited
A. Vahdat and D. Becker, Epidemic routing for partially-connected ad A

hoc networks, in Tech. Rep. CS-200006, Duke University, 2000.
61
T. G. Kurtz, Solutions of Ordinary Differential Equations as Limits of Pure Jump Markov Processes, Journal of Applied Probability, vol. 7, no. 1,
pp. 4958, 1970.
M. Benam and J.-Y. Le Boudec, A class of mean eld interaction models for computer and communication systems, Performance Evaluation,
vol. 65, no. 11-12, pp. 823838, 2008.
62
pp. 4958, 1970.
vol. 65, no. 11-12, pp. 823838, 2008.
63
pp. 4958, 1970.
vol. 65, no. 11-12, pp. 823838, 2008.
64
pp. 4958, 1970.
vol. 65, no. 11-12, pp. 823838, 2008.
65
Zhang, X., Neglia, G., Kurose, J., Towsley, D.: Performance Modeling of Epidemic Routing. Computer Networks 51, 28672891
(2007)
66
Accuracy of the modeling
(2007)
67
without any limitation on the number of copies !

(2007)
68
Outline
I.

1.
2.
II.
Introduction
1.
2.

1.
2.
3.
3.

1.
2.
3.
4.

mobility

69
Spray-and-Wait routing
- Goal: control the total number of disseminated copies thanks to a counter per packet
- i hands floor(c/2) forwarding tokens to j and keeps the rest ceil(c/2) for itself
1D
2D
1D
B
1D
E
5D
D
2D
1D
3D
1D
The counter does not allow for

more than L=5 copies to be
disseminated
C
B
T. Spyropoulos, K. Psounis, and C. Raghavendra, Efficient routing in intermittently connected mobile networks: the multi-copy case,
IEEE/ACM Trans. Netw., vol. 16, no. 1, pp. 7790, Feb 2008.
70
Spray-and-Wait versus Epidemic

Epidemic routing:
no limit on the number of transmissions ( nb of pkts . N)
mean time for delivery of one packet: log2(N)
Spray-and-Wait:
number of transmissions nb of pkts . L
When all nodes move in an IID manner, for any given max
number of copies L, Spray-and-Wait minimizes the expected
delivery time.
T. Spyropoulos, K. Psounis, and C. Raghavendra, Efcient routing in intermittently connected mobile networks: the multi-copy case,
IEEE/ACM Trans. Netw., vol. 16, no. 1, pp. 7790, Feb 2008.
71
Outline
I.

1.
2.
II.
Introduction
1.
2.

1.
2.
3.
3.

1.
2.
3.
4.

mobility

72
Why measuring human mobility?

Mobility increases the capacity of dense
mobile networks [Grossglauser and Tse]
May also create disconnections in sparse
MANETs
But: Human mobility patterns determine
communication opportunities
73
Duncan Watts (after Milgrams

experiment)
Six degrees - the science of a connected
age, 2003
Six degrees of separation is the idea that everyone is

on average approximately six steps away, by way of
introduction, from any other person in the world, so
that a chain of "a friend of a friend" statements can
be made, on average, to connect any two people in
six steps or fewer.
J. Vaucher, Small World Networks, April 2005
74
Understanding why to leverage this feature of

social networks
Why are there short chains of acquaintances
linking together arbitrary pairs of strangers?
In a random network, if everybody has 100
friends distributed randomly in the world
population, this isnt strange.
In 6 hops, you can reach 1012 people > 6.109
(world pop.)
BUT: our social networks tend to be clustered.
J. Vaucher, Small World Networks, April 2005
75
Random-graph theory
In mathematical terms a network is represented by a
graph. A graph is a pair of sets G(V,E), where V is a set
of N vertices V1, V2, ..., VN and E is a set of edges.
The theory of random graphs was introduced by Paul
Erdos and Alfred Renyi (1959-1961) after Erdos
discovered that probabilistic methods were often
useful in tackling problems in graph theory.
Problem: generating random graphs close to those
arising in the real-world (same features)
R. Albert and A.-L. Barabasi: Statistical mechanics of complex networks, Rev. Mod. Phys., Vol. 74, No. 1, January 2002
76
Some network measures

Characteristic Path Length (L) :
the average length of the shortest path connecting each
pair of nodes
Diameter: maximum value of path length

Clustering Coefficient (C) is a measure of local
interconnection
if node i has ki immediate neighbors, Ci, is the fraction of
the total possible ki*(ki-1) / 2 connections that are realized
between i's neighbors. C is just the average of the Ci's.
Clustering measures the fraction of neighbors of a node
that are connected themselves
77
The Erdos-Renyi model (1959)
N nodes, any two nodes are connected by an edge with probability p
Random-graph theory studies the properties of the probability space associated

with graphs with N nodes as N. Many properties of such random graphs can
be determined using probabilistic arguments.
Among the questions addressed by Erdos and Renyi, some have direct relevance to
an understanding of complex networks:
Is a typical graph connected?

How does its diameter depend on its size?
78
Social networks
Most of our friends come from our
geographical or professional neighborhood.
+ Our friends tend to have the same friends
--> High clustering coefficient
But also
There seems to exist short paths between any
pair of nodes
--> Low diameter
79
How to model real networks?
Regular Graphs have a high clustering coefficient

but also a high diameter
Random Graphs have a low diameter

but a low clustering coefficient
--> Combine both to model real networks: the Watts and Strogatz model
Random Graph (k=4)

Short path length
Regular Graph (k=4)

Long paths
L=logkN
L = n/(2k)
Almost no clustering
Highly clustered
C=k/n
C=3/6
Regular ring lattice

80
Small-World Networks
Random rewiring of regular graph

(by Watts and Strogatz)
With probability p rewire each link
in a regular graph to a randomly
selected node
Resulting graph has properties,
both of regular and random graphs
--> High clustering and short path
length
81
The small-world feature of human

mobile networks
Opportunistic networks exploit human mobility and
consequent device-to-device contacts to disseminate
content.
Human social networks are known to be clustered: we are
more likely to meet and spend time with our immediate
social circles such as family, friends or coworkers while we
meet infrequently with more casual acquaintances and
accidentally with strangers.
However, despite this high clustering, any pair of nodes is
typically separated by a surprisingly small number of hops:
known as the small-world phenomenon.
These properties have been shown to appear in
opportunistic mobile networks too.
82
Mobility datasets of contact traces

Denition: a contact is the period of time during which two devices are in
mutual radio transmission range and can exchange data.
Some traces (more on crawdad):
WLAN Access Point associations from the Dartmouth and ETH Zurich campus
Bluetooth contacts from the MIT
A contact trace file is made of:

[Time stamp] [Node1 ID] [Node2 ID] [Duration]
83
From the contact trace to the contact

graph
For each pair of nodes (i,j), from (frequency,duration), a
weight wij is extracted
T. Hossmann, T. Spyropoulos, F. Legendre, A complex network analysis of human mobility, NETSCICOM 2011
84
The small-world feature of human

mobile networks
--> The small-world behavior, typical for social

networks, is also observed in the network of
physical encounters.
T. Hossmann, T. Spyropoulos, F. Legendre, A complex network analysis of human mobility, NETSCICOM 2011
85
Not IID mobility anymore: impact on the

inter-meeting times?
IID mobility:
Inter-meeting times mobile/mobile have shown to
follow an exponential distribution
Groenevelt et al.: The message delay in mobile ad hoc networks. Performance Evaluation, 2005
Human mobility:
- Inter-meeting times have shown
to follow power law distribution
CCDF: Pr{ X > x } = (x/d)
86
Chaintreau et al.: Impact of human mobility on the design of opportunistic forwarding algorithms. Infocom, 2006
Outline
I.

1.
2.
II.
Introduction
1.
2.

1.
2.
3.
3.

1.
2.
3.
4.

mobility

87
Routing in heterogeneous mobile

networks: Utility-based routing
In PSN, to overcome the problem of sporadic and short-lived connectivity, opportunistic

routing schemes have been proposed such that:
a message may be stored and carried by a node for long periods of time until a
communication opportunity arises (mobility-assisted)
local forwarding decisions are made independently, and often with little or no
knowledge about the destinations position, with the goal that a message will eventually
be delivered (opportunistic)
multiple copies of the same message may be propagated in parallel (replication), to
increase the probability of at least one being delivered.
We have seen: two-hop routing, epidemic and spray-and-wait under IID mobility
But: nodes are heterogeneous in terms of their ability to deliver a message,
--> greedy replication may fail to discover the better relays
Given a xed budget of message copies, how can we best allocate it to a network consisting
of nodes with varying capabilities and behaviors?
--> instead of naively (greedily) handing copies to the rst nodes encountered, choose relays
according to some utility function: utility-based replication
88
Utility-based routing
Problem: in a heterogeneous environment, where a limited budget of L message

copies needs to be distributed to L relays, a mechanism is necessary to distinguish
the better relays and avoid using the least useful ones (given some optimization
criterion).
However, this problem is not trivial, in particular because candidate relays are
encountered in an online fashion.
Utility-based Spraying:
each node i maintains a utility function Ui(j) for every other node j in the network.
Ui(j) reects the probability that node i will deliver a message to node j , and it may be
based on a number of different parameters (e.g. encounter history, mobility, friendship,
etc.)
if a node i (either the source or a relay) carrying a message copy for a destination d and
having a forwarding counter c >1, meets node j with no copy of the message, the
message is copied to j if:
rule 1: if Uj(d) > Uth for some Uth threshold
rule 2: if Uj(d) > Ui(d)
i hands floor(c/2) forwarding tokens to j and keeps the rest ceil(c/2) for itself
T. Spyropoulos, T. Turletti, and K. Obraczka, Routing in delay-tolerant networks comprising heterogeneous node populations, IEEE Trans. on
Mob. Comp., vol. 8, pp. 1132 1147, 2009.
89
Some utility functions

Last-Seen-First (LSF) Spraying
Choose as relays the nodes that have seen the destination most recently.
Each node i maintains a timer i(j) for every other node j in the network,
which records the time elapsed since the i and j last encountered each other
as follows: initially set i(i) = 0 and i(j) = .
If i encounters j, set i(j) = j(i) = 0; otherwise increase each i(j) at every time
unit.
Finally Ui(j) = 1/(1+ i(j))
Most Mobile First (MMF) Spraying:

Assume that each of these nodes carries a label that states the type of the
node, e.g. BUS,TAXI, PEDESTRIAN, BASE STATION.
Assume there are m total node labels.
Dene the utility of node i as Ui(j) = Ui = LABEL(i).
Labels are put in a preference order: LABEL1 < LABEL2 << LABELm
Mob. Comp., vol. 8, pp. 1132 1147, 2009.
90
Some utility functions

Most Social First (MSF):
Define the sociability of a node:
Maintain a running average of its perceived

sociability index at a node:
Mob. Comp., vol. 8, pp. 1132 1147, 2009.
91
An example of exploiting the social

clustering into communities: BubbleRap
The same idea: as mobility is often unpredictable in
PSN, and topology changes can be rapid, rather than
exchanging much control traffic to create unreliable
routing structures, search for some characteristics of
the network which are less volatile than mobility.
If we can detect these social mobility patterns online
in a decentralized way, we can put the algorithms
into practical applications.
P. Hui, J. Crowcroft, and E. Yoneki, Bubble rap: Social-based forwarding in delay tolerant networks, IEEE Trans. on Mob. Comp., vol. 10, no.
11, pp. 1576 1589, Nov 2011.
92

Combines the knowledge of community structure
with the knowledge of node centrality to make
forwarding decisions.
Each node belongs to at least one community.
Each node has a global ranking (i.e. global centrality)
across the whole system, and also a local ranking within its
local community.
11, pp. 1576 1589, Nov 2011.
93

11, pp. 1576 1589, Nov 2011.
94

Rank is the betweeness centrality: number of times a node acts as a relay
for other nodes on all the shortest delay paths.
- small number of nodes which have

extremely high relaying ability,
and a large number of nodes have
moderate or low centrality values,
- human heterogeneity: people differ in
their popularity.
For instance, a salesperson or politician interacts with many
others, making themselves highly-ranked nodes in our graph,
compared to (say) the average computer scientist.
--> Choose popular hubs as relays

11, pp. 1576 1589, Nov 2011.
95
Estimating the rank online

The total degree (unique nodes seen by a node throughout
the experiment period) is not a good approximation of the
node centrality.
Instead the degree per unit time (for example the number of
unique nodes seen per 6 hours ) and the node centrality have
a high correlation value.
More generally: how to consider that there is an edge
between two nodes in the social graph, in terms of the
number, frequencies or duration of the meetings between
those 2?
96
Outline
I.

1.
2.
II.
Introduction
1.
2.

1.
2.
3.
3.

1.
2.
3.
4.

mobility

97
Inferring human communities

Problem 1: from the contact trace to the contact graph
Community detection :
Many community detections algorithms: Girvan-Newman
(WNA), modularity maximization, Louvain, clique-based
K-CLIQUE allows for overlapping communities, but was
designed for binary graphs --> problem of turning the
mobility trace into a binary graph
Weighted Network Analysis (WNA) can work on weighted
graphs directly, but it cannot detect overlapping
communities.
Problem 2: decentralized processing (see ref)
P. Hui, E. Yoneki, S. Yan Chan, and J. Crowcroft. Distributed community detection in delay tolerant networks. In Proceedings of ACM/IEEE
MobiArch 2007.
98
Example of K-clique community detection
A k-clique is a completely connected subgraph of k nodes.

2 k-cliques are said to be adjacent if they share k 1 nodes
A k-clique community is an union of all k-cliques that can be reached from each other
through a series of adjacent k-cliques.
99
From the contact trace to the contact

graph
Aggregate the entire sequence of contacts of a trace to a
static, weighted contact graph G(N,W) with weight matrix W =
{wij} : strength of the relationship between nodes i and j.
Various metrics possible: age of last contact, contact
frequency or aggregate contact duration.
100
Density-based Aggregation
A more useful and robust approach than time-based aggregation, can be to choose
the aggregation function such that the resulting social graph has a given density:
If we want to operate the social graph at a certain density, say, d(Gn) = 0.2, we
choose the best edges, according to some criterion, such that En will have the
desired cardinality:
Most Recent Contacts
Most Frequent Contacts
T. Hossmann, F. Legendre, T. Spyropoulos, From Contacts to Graphs: Pitfalls in Using Complex Network Analysis for DTN Routing, in IEEE
NetSciCom 09, Rio de Janeiro, Brazil, April, 2009
101
Density-based Aggregation online

An online algorithm that uses concepts from
unsupervised learning and spectral graph
theory to infer this correct graph structure.
Allows each node to locally identify and adjust
to the optimal operating point.
T. Hossmann, F. Legendre, T. Spyropoulos, From Contacts to Graphs: Pitfalls in Using Complex Network Analysis for DTN Routing, in IEEE
NetSciCom 09, Rio de Janeiro, Brazil, April, 2009
102
Conclusion
A lot of other community-based routing algorithms for
mobile social DTN (PSN): SimBet, PeopleRank, etc
Some other problems to overcome:
What are the contacts that matter: are they regular contacts, or
random contacts?
Considering content caching, what are the relays the content
should be placed to?
What social features to consider to better match social mobility?
What do we put inside social information: needs to
understand what type of social information is the most relevant
for forwarding, and analyze the impact of choosing a good and
meaningful social information on social forwarding decisions.

103
Outline
I.

1.
2.
II.
Introduction
1.
2.

1.
2.
3.
3.

1.
2.
3.
4.

mobility

104
Efficient Network Coded Data

Transmission in DTN
Yunfeng Lin, Baochun Li, Ben Liang, "Efficient Network Coded Data Transmissions in Disruption Tolerant Networks," in the
Proceedings of IEEE INFOCOM 2008, Phoenix, Arizona, April 2008.
Yunfeng Lin, Baochun Li, Ben Liang, "Stochastic Analysis of Network Coding in Epidemic Routing," in IEEE Journal on Selected
Areas in Communications, Special Issue on Delay and Disruption Tolerant Wireless Communication, Vol. 26, No. 5, pp. 794-808,
June 2008.
105
Motivation Constraints in DTN
Opportunistic connections between nodes

Nodes have limited transmission capabilities
Buffer space limitations
Battery power limited
Nodes are mobile
Delay in packet delivery will be large
Node density is low
106
by Yunfeng Lin
Network model
Settings:
only a single unicast session
: average number of meetings a node has per time unit
sparse DTN: =N remains constant as N increases
--> the network is connectivity-limited
mobility model: fast and uniform (RW, RWP,)
--> routing strategies must permit timely delivery of information to a certain

destination with high probability: use of replication
replication leads to energy and memory consumption
finite duration of radio contacts --> file split into packets
Objective: optimize the file transfer from S to D by minimizing both its
delay, the memory and energy required by the store and forward process
107
Motivation Binary Spraying vs. ER

Epidemic routing:
Forwarding is not limited
Spray-and-Wait: example with L=3
C
D
D
D
There are already 3 copies,
no more forwarding
B
A
108
Binary Spraying Vs ER (contd)

Epidemic routing:
no limit on the number of transmissions ( nb of pkts . N)
mean time for delivery of one packet: log2(N)
Spray-and-Wait:
number of transmissions nb of pkts . L
mean time for delivery of one packet: log2(L)+N/L
109
Motivation NC Vs Replication
b a
D cannot recover a and b

b
1
D 110
Motivation NC Vs Replication
a+b
b a
D can recover a and b

a+2b
2a+b
1
D 111
Protocol - Principle
This protocol called the E-NCR, is a combination of Network coding and
Binary spraying.
ER
NCER
Binary Spraying
E-NCR
NCER Network Coding based Epidemic routing

ER Epidemic Routing
E-NCR - Efficient Network coding based routing
112
by Yunfeng Lin
Protocol - Assumptions
There is one source S with K info packets to be transmitted, n relay nodes
and a destination D
For every opportunistic contact, only one packet can be transmitted.
Relay nodes have buffer space B, defined as 1 B K
No other background traffic
A packet in the buffer of a node is purged as soon as an ACK is received
from D or the Time-to-live field reaches zero.
Buffer structure:
pkt_index
pkt_counter
pkt_content
113
by Yunfeng Lin
K=2
K=3
L=7
E-NCR: an example
Time
0
1
Node 1
Node 2
Buffer content Node 1
R1
R2
Buffer content Node 2
d=a+b
e=2a+3b
f=a+2b
R1
R2
R2
R3
pkt_index
pkt_counter
pkt_content
S
S
Buffer structure:
3d+5e
114
Protocol - Description
SOURCE-RELAY:
K' = K + some more encoded packets

L = c * log k, where c is some constant
i = 0;
S = K';
do
{
if(detect any node and <i,l> not already there with that node)
{
send an encoded packet <i, L, co-efficients, packet>
i++;
}
}while(S != i);
115
by Yunfeng Lin
RELAY-RELAY, SENDER SIDE:
do
{
if(detect any node X)
{
get spray list of X; //list element is a tuple <i, l>,where i is index of packet, l is the
//remaining spray count
do
{
compare this->spraylist with x->spraylist;
if(any this->spraylist-><i, l> such that l >=0 and i does not exist in x->spraylist)
{
send encoded packet <i, floor(l/2)> to node x;
update tuple <i, l> to <i, ceil(l/2)>;
}
}while(end of x->spraylist);
}
}while(true);
116
by Yunfeng Lin
RELAY-RELAY, RECEIVER SIDE:
DESTINATION:
if(packet received)
{
if(buffer size == max_buffer_size)
{
encode incoming packet with all
packets in list;
}
else
{
place packet in free slot;
}
add <i,l> of incoming packet to spray
list;
}
do
{
if(got a packet)
{
add to packet list
try to decode list of packets;
if(decode possible)
{
exit loop;
}
}
}while(true);
117
by Yunfeng Lin
Performance
118
by Yunfeng Lin
Some Limitations
Destination has to wait till minimum of K
encoded packets are received
Some packets which have linear dependence
could arise during encoding at relays.
119
by Yunfeng Lin

coursWN1 Lucile2014

Uploaded by

Copyright:

Available Formats

coursWN1 Lucile2014

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

coursWN1 Lucile2014

Uploaded by

Copyright:

Available Formats

Course: Content Distribution in

Ubiquitous / Pervasive / Mobile

Another term: pervasive computing.

A. Heinemann, Collaboration in Opportunistic Networks, PhD dissertation, Darmstadt 2007

Opportunistic routing in MANET

Network coding: beyond routing

Motivation for opportunistic networking

Routing in heterogeneous DTN: mobile social networks

Some uniform mobility models and their characterization

Network coding under Spray-and-Wait in DTN

Oriented valued graph G = (V, E, c)

Value of the flow

Problem of max flow

(X, Y) is a cut of G iif:

Capacity of the cut:

X = {A,B,C,D} Y = {E,F} c(X,Y) = 14 f (X,Y) = 12

->Theorem of min-cut max-flow:

Introduction: multicast with routing

Introduction: multicast with network

With network coding,

Linear Network Coding

Y (e3 ) 1,e3 X (v,1) 2,e3 X (v,2) 3,e3 X (v,3)

v1e1 ,e5 ee1 ,4e4 e4e,e66

Z (v4 ,2) e5 , 2Y (e5 ) e6 , 2Y (e6 ) e7 , 2Y (e7 )

Z (v4 ,3) e5 ,3Y (e5 ) e6 ,3Y (e6 ) e7 ,3Y (e7 )

3,ee ,1 3,ee , 2 3,ee ,3

Network Coding Solution

A 2,e 2,e 2,e ,

Connection between an Algebraic

Finding Network Coding Solution

Random Network Coding

On lossy channels: NC for unicast

To cope with erasures on a channel with , we need to send N=K+M pkts

On lossy channels: NC for unicast

RLNC at relay: R generates pkts on the fly by mixing those

Inter-session network coding

Inter-session network coding

The COPE example

improves the throughput of

The butterfly example

Network coding: beyond routing

Motivation for opportunistic networking

Routing in heterogeneous DTN: mobile social networks

Some uniform mobility models and their characterization

Network coding under Spray-and-Wait in DTN

XORs in The Air: Practical Wireless

How to optimize throughput?

An information exchange scenario

Multi-hop unicast requires 4 transmissions

Can Network Coding help? - An idea

The COPE approach

Opportunistic coding using local info

Emulation and testbed results

COPE: Opportunistic Coding Protocol

Checks packets in queue

Combines packets traversing the same three hops in

Metadata in a header between MAC and IP

Broadcast encoded packets