1

Module I

Routing in mobile ad hoc networks


Prof. JP Hubaux





Mobile Networks
http://mobnet.epfl.ch
2
The classical solution for mobile networks
2
nd
generation (GSM, IS-41,) and 3
rd
generation (UMTS,)
deployed soon
Huge, expensive fixed infrastructure
License for a share of the spectrum
Operational responsibility: network operators (telcos, ISPs)
3
The new paradigm: ad hoc networks
Terminal and node merge
Everything is potentially mobile
Initial applications: communication in the battlefield (Packet
Radio Networks, in the 70s)
The network is self-organized when it is run by the users
themselves
Similar trend at the application layer: peer-to-peer
(e.g., Napster Gnutella)
4
Application examples of ad hoc networks
Sensor networks
Hybrid cellular / ad hoc networks (multi-hop cellular
networks)
Cars
Assisted driving (adaptive cruise control,)
Collision avoidance
Optimization of traffic flows

Crisis networks (e.g., rescue operations after major
disaster)
Military networks
5
Upper bound for the throughput of ad hoc
networks
If we have:
- identical randomly located nodes
- each capable of transmitting bits/s
Then the throughput ( ) obtainable by each node
for a destination is
( )
log
n
W
n
randomly chosen
W
n
n n

| |
= O
|
|
\ .
Ref: P. Gupta, P. Kumar, The Capacity of Wireless Networks
IEEE Transactions on Information Theory, March 2000
6
Intuition behind the upper bound
N nodes (users)
O(N) users O(N) users
Cut set ~ N
O(N) transmissions from left to right
over
O( ) transmission links
mean
O( ) capacity per attempted transmission
N
1
N
Ways to improve scalability:
Directional antennas
Locality of the traffic
Hybrid system
7
Routing in ad hoc networks
Peculiarities
Node mobility
High rate of link failure
Traditional routing approaches are not well suited
Assumptions
Multihop communication
Symmetric links (in most cases)
Omnidirectional antennas (in most cases)
All nodes have equal capabilities and responsibilities
Figures of merit
Latency of route discovery
Overhead (bandwidth, energy, processing power)
Security
Current status of research:
Many, many proposals
Optimal solution depends on deployment scenario: mobility
patterns, radio model, traffic characteristics,

8
Brief reminder : Link-state protocols
Example: OSPF
May consume a lot of resources to update the routes
Techniques to alleviate the problem : limit the
propagation of information
Does not seem to be well suited to cope with mobility
9
Distance vector routing (1/2)
0 1 5

1 0 1 3
5 1 0 7

3 7 0
A
B
D
C
1
3
1
5 7
A
B
C
D
A B C D
1 0 1 3
Distance
vector
Distance
vector of B
2 1 2 4
+ Distance from A to B =
Cost to dest.
via B
Take the min
0 1 2,B 4,B
(1 row stored in each node)
10
Distance vector routing (2/2)
Even if the updates are asynchronous, the routing
tables converge
The algorithm is often called Bellman-Ford
Problem: undesirable behaviour when links go up
and down (e.g., count to infinity problem)


11
Routing protocols for wireless ad hoc networks
Response time,
bandwidth
Energy
Mobile ad hoc networks
Sensor networks
Proactive
protocols
Reactive
protocols
Destination-Sequenced
Distance-Vector (DSDV)
Optimized Link-
State Routing
(OLSR)
Ad Hoc On-Demand
Distance-Vector
(AODV)
Dynamic
Source
Routing
(DSR)
Geography-
based routing
Cluster-based
(or hierarchical)
routing
Geodesic
packet
forwarding
12
Dynamic source routing (DSR)
Reactive routing protocol
2 phases, operating both on demand:
Route discovery
Used only when source S attempts to to send a packet to
destination D
Based on flooding of Route Requests (RREQ)
Route maintenance
makes S able to detect, while using a source route to D, if it can
no longer use its route (because a link along that route no
longer works)
13
DSR: Route discovery (1)
E
G
M
H
R
F
A
B
C
I
D S
K
N
L
P
J
Q
14
DSR: Route discovery (2)
E
G
M
H
R
F
A
B
C
I
D S
K
N
L
P
J
Q
(S)
15
DSR: Route discovery (3)
E
G
M
H
R
F
A
B
C
I
D S
K
N
L
P
J
Q
(S,A)
(S,E)
16
DSR: Route discovery (4)
E
G
M
H
R
F
A
B
C
I
D S
K
N
L
P
J
Q
(S,E,G)
(S,B,C)
17
DSR: Route discovery (5)
E
G
M
H
R
F
A
B
C
I
D S
K
N
L
P
J
Q
(S,E,G,J)
(S,A,F,H)
18
DSR: Route discovery (6)
E
G
M
H
R
F
A
B
C
I
D S
K
N
L
P
J
Q
(S,A,F,H,K)
19
DSR: Route discovery (7)
E
G
M
H
R
F
A
B
C
I
D S
K
N
L
P
J
Q
(S,A,F,H,K,P)
20
DSR: Route discovery (8)
E
G
M
H
R
F
A
B
C
I
D S
K
N
L
P
J
Q
RREP(S,E,G,J,D)
21
DSR: Route Discovery (9)
Route reply by reversing the route (as illustrated)
works only if all the links along the route are
bidirectional
If unidirectional links are allowed, then RREP may
need a route discovery from D to S
Note: IEEE 802.11 assumes that links are
bidirectional
22
DSR: Data delivery
E
G
M
H
R
F
A
B
C
I
D S
K
N
L
P
J
Q
DATA(S,E,G,J,D)
23
DSR: Route maintenance (1)
E
G
M
H
R
F
A
B
C
I
D S
K
N
L
P
J
Q
DATA(S,E,G,J,D)
X
24
DSR: Route maintenance (2)
E
G
M
H
R
F
A
B
C
I
D S
K
N
L
P
J
Q
X
RERR(G-J)
When receiving the Route Error message (RERR),
S removes the broken link from its cache.
It then tries another route stored in its cache; if none,
it initializes a new route discovery
25
DSR: Optimization of route discovery: route
caching
Principle: each node caches a new route it learns by
any means
Examples
When node S finds route (S, E, G, J, D) to D, it also learns
route (S, E, G) to node G
In the same way, node E learns the route to D
Same phenomenon when transmitting route replies
Moreover, routes can be overheard by nodes in the
neighbourhood
However, route caching has its downside: stale
caches can severely hamper the performance of the
network

26
DSR: Strengths
Routes are set up and maintained only between
nodes who need to communicate
Route caching can further reduce the effort of route
discovery
A single route discovery may provide several routes
to the destination
27
DSR: Weaknesses
Route requests tend to flood the network and
generally reach all the nodes of the network
Because of source routing, the packet header size
grows with the route lengh
Risk of many collisions between route requests by
neighboring nodes need for random delays before
forwarding RREQ
Similar problem for the RREP (Route Reply storm
problem), in case links are not bidirectional

Note: Location-aided routing may help reducing the
number of useless control messages
28
Ad Hoc On-Demand Distance Vector
Routing (AODV)
As it is based on source routing, DSR includes
source routes in data packet headers
Large packet headers in DSR risk of poor
performance if the number of hops is high
AODV uses a route discovery mechanism similar to
DSR, but it maintains routing tables at the nodes
AODV ages the routes and maintains a hop count
AODV assumes that all links are bi-directional
29
AODV : Route discovery (1)
E
G
M
H
R
F
A
B
C
I
D S
K
N
L
P
J
Q
30
AODV : Route discovery (2)
E
G
M
H
R
F
A
B
C
I
D S
K
N
L
P
J
Q
Note: if one of the intermediate nodes (e.g., A)
knows a route to D, it responds immediately to S
: Route Request (RREQ)
31
AODV : Route discovery (3)
E
G
M
H
R
F
A
B
C
I
D S
K
N
L
P
J
Q
: represents a link on the reverse path
32
AODV : Route discovery (4)
E
G
M
H
R
F
A
B
C
I
D S
K
N
L
P
J
Q
33
AODV : Route discovery (5)
E
G
M
H
R
F
A
B
C
I
D S
K
N
L
P
J
Q
34
AODV : Route discovery (6)
M
D
K
L
P
J
E
G
H
R
F
A
B
C
I
S
N
Q
35
AODV : Route discovery (7)
M
D
K
L
P
J
E
G
H
R
F
A
B
C
I
S
N
Q
36
AODV : Route reply and setup of the forward
path
M
D
K
L
P
J
E
G
H
R
F
A
B
C
I
S
N
Q
: Link over which the RREP is transmitted
: Forward path
37
Route reply in AODV
In case it knows a path more recent than the one
previously known to sender S, an intermediate node
may also send a route reply (RREP)
The freshness of a path is assessed by means of
destination sequence numbers
Both reverse and forward paths are purged at the
expiration of appropriately chosen timeout intervals


38
AODV : Data delivery
M
D
K
L
P
J
E
G
H
R
F
A
B
C
I
S
N
Q
Data
The route is not included in the packet header
39
AODV : Route maintenance (1)
M
D
K
L
P
J
E
G
H
R
F
A
B
C
I
S
N
Q
Data
X
40
AODV : Route maintenance (2)
M
D
K
L
P
J
E
G
H
R
F
A
B
C
I
S
N
Q
X
RERR(G-J)
When receiving the Route Error message (RERR),
S removes the broken link from its cache.
It then initializes a new route discovery.
41
AODV: Destination sequence numbers

If the destination responds to RREP, it places its
current sequence number in the packet
If an intermediate node responds, it places its record
of the destinations sequence number in the packet
Purpose of sequence numbers:
Avoid using stale information about routes
Avoid loops (no source routing!)
42
AODV : Avoiding the usage of stale routing
tables
S D A
B
DSN(D) = 5
: Forward path
S
D
A
B
DSN(D) = 5


DSN(D) = 8
1.
2.
S
D
A
B
DSN(D) = 5

DSN(D) = 8
3.
RREQ
S
D
A
B
DSN(D) = 5

DSN(D) = 8
RREP
4.
43
AODV : Avoiding loops
A B S D
C
: Forward path
Assume there is a route between A and D; link S-D breaks; assume A is not aware of this, e.g. because
RERR sent by S is lost
Assume now S wants to send to D. It performs a RREQ, which can be received by A via path S-C-A
Node A will reply since it knows a route to D via node B
This would result in a loop (S-C-A-B-S)
The presence of sequence numbers will let S discover that the routing information from A is outdated
Principle: when S discovers that link S-D is broken, it increments its local value of DSN(D). In this way,
the new local value will be greater than the one stored by A.
X
44
AODV (unicast) : Conclusion
Nodes maintain routing information only for routes
that are in active use
Unused routes expire even when the topology does
not change
Each node maintains at most one next-hop per
destination
Many comparisons with DSR (via simulation) have
been performed no clear conclusion so far
45
Geodesic Packet Forwarding
L. Blazevic, S. Giordano, J.-Y. Le Boudec (IP4)
AP -geographical anchor point
AGPF (Anchored Geodesic Packet Forwarding) - source
routing with anchors

S
AP1
D
AP2
S has anchored path {AP1,AP2}
from S: packets are forwarded in direction of AP1
from A: packets are forwarded in direction of AP2
from B: packets are forwarded in direction of Ds position
from X : use of Terminode Local Routing (TLR)
A
B
X
TLR area of X
46
Other (Swiss) proposals
Last Encounter Routing
H. Dubois-Ferrire, M. Grossglauser, M. Vetterli (EPFL, IP1 & 7)
Principle: Nodes exchange information about their previous
encounters
No explicit location service, no transmission overhead to to
update the state
Ongoing work: prediction, based on declared mobility
Face routing
F. Kuhn, R. Wattenhofer, A. Zollinger (ETHZ, IP9)
Principle: exploit the geometric properties of the connectivity
graph
Worst-case optimal


47
Conclusion on routing
DSR and AODV are the mainstream proposals
Both have been extensively studied (by simulation)
No clear superiority of one wrt the other
Scalability is still an open issue
Other very promising proposals
48
References
Ch. Perkins: Ad Hoc Networking, Addison Wesley,
2001
www.mics.org: IP1, IP4, IP7 and IP9