Routing in Wireless Ad Hoc Networks

7/27/2019 Routing in Wireless Ad Hoc Networks

1/37

1

Routing in WirelessAd Hoc Networks

3

MANETs

Nodes are computing and communicationdevices, e.g laptops, PDAs, mobile phones,even sensors

Nodes organize and maintain the networkby themselves

A node is both a host and a router IP is used at the network layer First and foremost issue: routing

4

Problems with Routing

1.Self-organizing networks Need for distributed algorithms

2.Topology changes dynamically Mobile nodes (joining in or leaving) Unannounced loss of network connectivity due

to the time-varying channel nature

3. Link failure / repair Network partitions Loop formation during temporary node failures


2/37

2

5

Problems with Routing

4. Asymmetric links: links may be unidirectional

5. Limited bandwidth Protocols using flooding create high traffic

and control overhead

8. New performance criteria

Number of hops (delay) Available Bandwidth Route stability despite mobility Energy consumption

6

Routing Protocols for MANETs

Find/maintain a route provided itexists

Another problem is to ensure that aroute exists, e.g., through power control

[Ramanathan00,Wattwnhofer00]

Assume that nodes are willing tocooperate

7

Topology based routing Proactive approach, e.g., DSDV, OLSR Reactive approach, e.g., DSR, AODV Hybrid approach, e.g., Cluster, ZRP

Routing Protocols for MANETs

Position based routing Location Services: e.g., DREAM, Quorum-

based, GLS, Home zone etc.

Forwarding Strategy: e.g., Greedy, GPSR,RDF, Hierarchical


3/37

3

8

The Proactive Approach

Based on distance vector and link-statemechanisms

Attempts to maintain consistent, up-to-daterouting information from each node to everyother node in the network

Respond to changes by propagating updatesthroughout the network

Good for connection-less traffic where youmay send traffic to any node at any time

9

The Reactive Approach

Routes are only created when desired by thesource node

Route discovery protocol Route discovery ends either when a route is

discovered or all possible communication

paths have been examined

Once a route is established, its maintained bysome sort of route maintenance protocol

10

Trade-off

Reactive more suitable for a mobileenvironment with limited bandwidth

Proactive preferred when time constraintsare important

ProactiveApproach

ReactiveApproach

Latency Low High

Overhead High Low


4/37

4

11

Proactive Protocols

DV for fixed networks, DSDV forwireless networks

12

Distance Vector Routing

Route discovery algorithm used by RIP(DVR does not address routemaintanance)

Based on Bellman-Ford or Ford-Fulkersonalgorithms but DVR does not require ana-priori knowledge of the network

13

Distance Vector Routing (1)

Each node has a routing table where distanceto all reachable destinations are recorded

Each node is responsible for keeping trackand informing its neighbors of its distanceto each destination (main idea at the basisof distance vector-based protocols)

A router computes its distance to a destinationbased on its neighbors distance to thedestination


5/37

5

14


A node must know its own ID and the cost of its links toeach neighbor (the cost can be expressed in terms ofnumber of nodes, bandwidth, latency, )

1. Initially, each node sets the cost of the link to itself to0 and to any other node to infinity

2. A node broadcasts its routing table3. A node, upon receiving information from neighbors,

computes minimum cost-path toward destination andmakes changes to its routing table if needed

4. Either periodically or every time a node updates itstable, it broadcasts it to neighbors.

The algorithm converges by iteration

15


At node i, the minimum cost path betweennodes i andj is computed as

where:

is the path cost at time step n

is the cost of the link between iandits neighbor k

Dn+1

(i, j) =min Dn(i, j),min

kd(i,k)+D

n(k, j){ }

16

Distance Vector Routing forAddress D

R

12

3

45

172

35

5

41

Link cost

Link number


6/37

6

17

Distance Vector Routing for

Address D

R

12

3

4

5

97

62

11829

81

172

35

5

41

Cost fromneighbor todestination D

18


R

12

3

4

5

98

99

97

123 70

Cost for Rto get to Dvia this link

Minimum

cost route

17

2

35

541

97

62

118 29

81

19


R

70

70

7070

70

Cost fromR to D12

3

4

5

17

2

35

5

41


7/37

7

20

Problems with Distance Vector

1. Does not scale well with the number of nodesin the network (overhead O(n2))

2. Slow convergence to the lowest cost route3. Slow recovery time if there are link failures,

hence routing problems during disruptiontimes

Count-to-infinity: Router loops may takeplace when degradation of a link cost occurs

Solved only when the cost of thealternative route including a loop reachesthe cost of the degraded link

21

Count-to-Infinity: An Example

A

B

A

B1

2

3

2 A

B5

4 A

B7

6

Node D becomes unreachableD

Cost of the link A-B and B-D equal to 1 =>cost of the path between A and D equal to 2

Route cost to D

21

It occurs only if A sends its update before B

22

DSDV (1):Destination SequencedDistance Vector Routing

Proposed by Perkins [94], based on Bellman-Ford routing mechanism

Each node maintains a routing table thatrecords all possible destinations

Each table entry contains1. Destination node ID (i.e., IP address)2.Next hop3.Number of hops to the destination4.A sequence number (SN) is used to

distinguish old vs. new routes and avoidloops


8/37

8

23

DSDV (2)

Routing tables are periodically transmitted tomaintain table consistency

A full dump record reports on the entirerouting table. Lots of traffic, since eachtable is probably larger than the networkprotocol data unit

Incrementalupdates provide information onchanges performed since last full dump

A new route broadcast contains The address of D The number of hops to reach D The SN of the route SN unique to this broadcast

24

DSDV (3)

Every change is broadcast to the neighbors(incremental update)

Upon reception of a route update:If the node doesnt have such information, it adds

an entry to its table

Otherwise, it checks the SN and updates the tableonly in case of fresher information

The route with most recent SN is used If two routes have the same SN, the route with the

smaller metric (== shorter route) is used

When a link fails, an metric is used and the route SNis increased

25

MH4

ForwardingTable

MH5

MH1

MH8

MH7


9/37

9

26

34

5

62

1

8

7

050-83MH6MH8

128-72MH6MH7

076-61MH6MH6

392-52MH6MH5

710-40MH4MH4

564-32MH2MH3

128-21MH2MH2

516-12MH2MH1

SNCostNextDest

MH4Forwarding

Table

27

MH6

MH3MH4

MH5

MH2

MH1

MH8

MH7

MH1 moves

MH6

MH1

28

MH3MH4

MH5

MH6

MH2

MH1

MH8

MH7

New MH4

ForwardingTable


10/37

10

29

New MH4Forwarding

Table

050-83MH6MH8

128-72MH6MH7

076-61MH6MH6

392-52MH6MH5

710-40MH4MH4

564-32MH2MH3

128-21MH2MH2

518-13MH6MH1

SNCostNextDest

34

5

62

1

8

7

30

DSDV (3)

Every change is communicated to the neighbors(incremental update)

Upon reception of a route update:If the node doesnt have such information, it adds

an entry to its table

Otherwise, it checks on the SN and updates thetable only in case of fresher information

The route with most recent SN is used If two routes have the same SN, the route with the

smaller metric (== shorter route) is used

When a link fails, an metric is used and theroute SN is increased

31

Count-to-Infinity: An Example

A

B

A

B1

2

3

2 A

B5

4 A

B7

6

Node D becomes unreachableD

Cost of the link A-B and B-D equal to 1 =>cost of the path between A and D equal to 2

Route cost to D

21

SN of the route notified by A is older than theone of the route declared invalid by B -> As infois not considered. B sends an update to A, and Atoo declares the route to D (through B) as invalid

NewSN


11/37

11

32

Detecting Link Failure

Detection: If data is flowing over a link, failure of

link layer to deliver a packet can be usedto assume link failure

If a node does not hear for a long time(how long?) from its neighbor

33

DSDV (4)

Nodes also keep track of the settling time ofroutes, i.e., the average time that routes to adestination will fluctuate before the routewith the best metric is received

A broadcast of a routing table will be delayed bythe length of the settling time

This avoids fluctuations in the table data andreduces the update packets

35

Advantages

Has moderate memory requirementsGuarantees loop-free paths (thanks toSNs)Simple update coordination protocolDisadvantages

Provides only a single path between each given S/D pairRequires selection of the periodic update interval,settling time and number of update intervals that shouldexpire before a route is considered stale: difficult toestimate

Update overhead: if full dump updates are used, overheadgrows as O(n2)

DSDV: Advantages andDisadvantages


12/37

12

36

OLSR

Optimized Link State Routing[Jacquet00] OLSR is based on the link state routing

approach

Traditional approach in Link State Routing: Every node generates control packets to

advertise its links status (i.e., its one-hopneighbors and the links cost)

Such information is propagated over thewhole network (flooding)

37

OLSROptimized Link State Routing[Jacquet00]

However, in OLSR: Sources build routes proactively by using

MultiPoint Relay nodes (MPRs)

Only MPRs need to generate and forward linkstate updates

OLSR thus leads to efficient flooding ofcontrol messages in the network: superfluousbroadcast packet retransmission as well asthe size of the link state packets are reduced

38

MPRs and Their Selection

Every node in the network selects its own MPR(s) A node selects as its multipoint relays those one-hop

neighbors that allows the node to reach its two-hopneighbors through symmetric links

Nodes exchange neighbor lists to know their 2-hopneighbors and link type, and choose the MPRs

No. of MPRs per node should be minimized


13/37

13

39

MPRs: Example (1)

Hp: all links are bi-directional

A

B F

C

D

E H

GK

J

E

Nodes C and E are multipoint relays of node A

40

MPRs: Example (2)

C and E forward information received from A Node E is a multipoint relay also for node H If the link H-J were unidirectional, then K would also be

selected as MPR by H

Note that C selects A as MPR, while E selects A and H

A

B F

C

D

E H

GK

J

Node that has broadcast state information from A

41

MPRs: Example (3)


14/37

14

42

Controlled Flooding of Link State Updates

24 retransmissionsto deliver a messageup to 3 hops

11 retransmissions todeliver a message upto 3 hops

43

Controlled Flooding of Link State Updates

OLSR is particularly suited for densenetworks

In sparse networks, every neighborbecomes a multipoint relay, then OLSR

reduces to pure link state protocol

44

How It Works: Hellos

Node N broadcasts Hello messages every Hellointerval to its 1-hop neighbors:

Hello message contains list of known 1-hopneighbors and

the link status (symmetric, asymmetric, or MPR) ofits 1-hop neighbors


15/37

15

45

How It Works: Neighbor Table

Node N builds a Neighbor Table that includes all its 1-hopneighbors and, for each of them, all 2-hop neighbors thatcan be reached through it. The link types are also recorded.

N selects its MPR nodes among its 1-hop neighbors suchthat it can reach all nodes that are 2-hops away throughsymmetric links

MPR election requires symmetric link to node NOnce N selects a neighbor, say X, as MPR, N tags its link

with X as MPR and, through the next Hello, N will notifyX about it

X maintains the list of nodes that selected itself as MPR

46

How It Works: Topology Control

MPR nodes broadcast Topology Control (TC) messagesevery TC interval to advertize link states, specifically

TC message contains list of 1-hop neighbors whohave selected this node as an MPR

Only MPR nodes can forward TC messages (butnote that still all network nodes receive them) ->

more efficient flooding

TC messages are used to build the Topology Table(and for routing table calculation afterwards)

47

Some Observations

For (any) control messages it is not requireda reliable transmission since they areperiodically sent

Each control message inlcludes a sequencenumber so that old messages with outdatedinformation can be detected and discarded


16/37

16

48

How It Works: Topology Table

Each node maintains a topology table where theinformation obtained from the received TCmessages is recorded

Each entry includes: Address of a potential destination (i.e., the

MPR selector in the received TC)

Address of the last hop to that destination(i.e., the MPR that generated that TC)

MPR selectors set sequence number

49

How It Works: Routing Table

Each node maintains a Routing Table where routes toany potential destination are recorded

The Routing Table is built based on the Topology Tableand the Neighbor Table

Each entry includes: Address of the potential destinationAddress of the next hop to reach that destination

Estimated distance to the destination Note that the Routing Table is re-calculated at every

change in either the Topology Table or the NeighborTable

50

Building the Routing Table (1)

Initialization:

Set h to 1; Use the Neighbor Table: take the 1-hop

neighbors (whose link status is not uni-directional) as destinations (i.e., h =1 ) andcreate the corresponding entry in theRouting Table;

set their distance to h = 1.


17/37

17

51

Building the Routing Table

Iterative procedure:

1) For each entry in the Topology Table:(i) if its destination does not correspond to anydestination in the Routing Table and(ii) if its last hop (i.e., MPR) is the destination of anentry in the Routing Table with distance equal to h, thenadd a new route entry in the routing table where:

dest = dest of the entry in the Topology Table; next-hop = next-hop of the route entry whose dest is

the last-hop in the Topology Table;

Set distance = h+1;3) Set h = h+1 and go to step 1 (till all entries in the

Topology Table have been processed)

52

Some Observations (1)

Hello interval impact on overhead vs. protocolreactivity:

Recall: Hellos are sent by all nodes to senseneighbors (but they are not rebroadcast)

TC interval impact on overhead vs. protocolreactivity:

Recall: TC messages are sent only by MPR nodesto advertize link state and allow routing tablecalculation

53

Some Observations (2)

Route are all composed of MPRs only (except sourceand destination): MPRs form a network backbone thatis in charge of routing all traffic

MPR nodes selection impact: How much more traffic must MPR nodes handle?

Higher load leads to higher energy consumption

Node mobility impact Consequences? Particularly for MPR nodes


18/37

18

54

Reactive Protocols

55

DSRDynamicSource Routing [Johnson96]

Entirely on demand: No periodic messages oradvertisements at any layer

Dynamic: it maintains a soft state, i.e., all state isdiscovered when needed and can be easily re-discovered if needed after a failure without impactingthe protocol functioning

Source Routing: the source specifies in the header ofeach data packet the entire route, not only the nexthop (no need for routing tables)

56

Assumptions

The network is fairly small (up to 200 nodes,i.e., network diameter of 10-15 nodes)

Each node maintains a cache containing allsource routes of which it is aware

Routes in the cache are continuouslyupdated as they are learned

Several routes can be cached to the samedestination

Either bi- or uni-directional links can be used


19/37

19

57

DSR: Two Phases

The protocol consists of two phases:

Route discovery: Started by S when S needs to send data to D

and doesnt have any route to D in cache

Route Maintenance: While using a route to D, S can detect if

the route is not longer valid and, in case, sendan error message

Upon route failure, S may use another route(if it knows it) or start a new route discovery

61

L

H

J

D

C

G

S

N

K

BA

Route Discovery

RREQ

Initiator ID

Initiator seq#

Dest ID

Partial route

TTL

Initiator seq#

Dest ID

Time to go

RREQ Table

If no reply by the time to go, S sends a new RREQtill a max no. of attempts have been made

Unique ID

QoS cost

62

L

H

J

D

C

G

S

N

K

B

A

Route Discovery

RREQrecently

seen?

DiscardRREQ

yes

no

myAddrin partialroute?

yes

no

store RREQID in list

decrease TTL andbroadcast packet

myAddr=target?

no

S

S

S,C

S,LS,C

S,CS,L

S,L

knownroute?

append myAddrto partial route

no


20/37

20

63

L

H

J

D

C

G

S

N

K

B

A

Route discovery succeeded !

Route Discovery

S,L,J

S,C,G

S,C,H

66

L

H

J

D

C

G

S

N

K

B

A

RREP

Route Reply

S,L,J,D

S,C,G,KS,C,G,K

S,L,J,D

S,L,J,D

S,L,J,D

Place route record andcost into RREP

(cost depends ontarget QoS metric)

RREP sent by reversingtheroute followed by RREQ. If

unidirectional, use any cachedroute to S. If none, send anRREQ and piggyback RREP

myAddr=target?

yes

67

L

H

J

D

C

G

S

N

K

B

A

DATA

Data Delivery

Write route inpacket header

and send packetto next hop

S,L,J,D

S,L,J,D

S,L,J,D

S,L,J,D


21/37

21

68

L

H

J

D

C

G

S

N

K

B

A

RREP

Route Caching

S,L,J,D

S,L,J,D

S,L,J,D

Place cached route andcost into RREP

(cost depends ontarget QoS metric)

RREP sent by reversingtheroute followed by RREQ. If

unidirectional, use any cachedroute to S. If none, send anRREQ and piggyback RREP

knownroute?

yes

69

Route Caching

Advantages of route caching Speeds up routing Reduces propagation of route requests

S caches the route to D contained in the RREP For each D, more than one route can be

cached, thus S can perform traffic routingover several possible routes

70

Route Maintenance

Along a source route, each node transmitting apacket is responsible for correct transmissionconfirmation

An ACK is needed for confirmationMAC or link-layer ACKs can be usedOverhearing next node forwarding the packet

(passive ACK: A hears B sending to C)

Use of DSR-ACKs: the node transmitting the packetcan require an ACK (which may follow a differentroute back)


22/37

22

72

Route Maintenance

S

BC

G

D

H

F

Mark link to nexthop as broken

ReceivedACK?

yes

Max#ACKreqexceeded

?

Remove fromcache all routes

including that link

Send an RERRto everysource that sent a packetto be routed over that link

no Route Cache (S)D: S,B,C,F,DD: S,B,C,G,H,DC: S,B,C

RERR

RERR

73

Route Maintenance

Nodes overhearing RERR remove fromtheir cache all routes including the brokenlink

S uses another route from its cache, if ithas any; otherwise it starts a new route

discovery The transport layer (e.g., TCP) should take

care of data retransmissions

74

Optimizations

Promiscuous mode: listening to arbitrary routes inuse (A hears B sending to C)

Replace with shorter routes / Store new routesBut promiscuous mode takes energy

Packet salvagingUpon link failureIf intermediate node has another route to D, it

replaces the original source

However, no double salvage is allowed !!!


23/37

23

75

Packet Salvaging: An Example (1)

Initial route: S (source) B C F D (destination) However, S and C have an alternative route to D in

their cache (S-B-Z-F-D and C-I-G-H-D, respectively)

At a certain time, F moves out of the network

SB

C

G

F D

HL

I

M

I

L

Z

76


Unfortunately, S receives no RERR, from either C orZ, then S keeps sending its packets toward C

C can salvage the packets by replacing the originalroute (S-B-C-F-D) in the packet with the new route(C-I-G-H-D)

SB

C

G

D

HL

I

M

I

L

Z

77


Now, assume the G-H link fails, then if G couldsave the packet again, a loop could be formed.Indeed (i) G does not know that the packet wasoriginated by S (it only sees C as originator), (ii)S could reply with the route S-B-Z-F-D

SB

C

G

F D

HL

I

M

I

REQREQ

REQ

REQ

REQ

REPREP

REP

REP

REP

L

Z


24/37

24

79

" Enrirely reactive" Soft state and support of unidirectional links" Source routing

" No need for routing decisions by intermediate nodes" Nodes can learn new routes from relayed packets" Guarantee that routes are loop-free

" Caching" Reduced route discovery overhead" One route discovery may yield many routes to D, dueto intermediate nodes replying from local caches

DSR: Advantages

80

" RREQ flooding may be huge" Possible collisions between RREQs propagated byneighboring nodes

" Contention between RREPs come back due to use oflocal caches (RREP Stormproblem)

" Packet delays/jitters due to on-demand routing" Headers may become too long with respect todata payloads -> suitable for small networks

" Cached routes may become invalid; stale cachescan adversely affect performance

DSR: Disadvantages

81

AODVAd-Hoc On Demand Distance Vector [Perkins99]

Builds routes on demand Attempts to improve DSR by using routing

tables at the nodes so that packets do nothave to include routes

Avoids loop formation by using sequencenumbers

Similar procedures of route discovery androute maintainance as in DSR


25/37

25

82

Assumptions

Each node maintains:

A broadcast ID to be used for RREQs Broadcast ID is incremented for each RREQ Combination of the node IP address & broadcast ID

uniquely identifies an RREQ

A (small) routing table with an entry for eachuseful destination node

83

Routing Table

A routing table entry includes:

Destination IP address Sequence number assigned to the path by the

destination (Destination Sequence Number, DSN)

Hop count Next hop List of nodes from which RREP msgs may be receivedand from which RREQ msgs are received Active route timeout (lifetime) Routing flag indicating whether the path is active or not

85

L

H

J

D

C

G

S

N

K

B

A

Route Discovery

RREQ

Initiator ID

Dest. Sequ. #

Dest ID

Broadcast ID

A new RREQ by node S for a destination isassigned a higher DSN (for the inverse routewith S as a destination)

TTL

QoS cost


26/37

26

86

L

H

J

D

C

G

S

N

K

B

A

Route Discovery

knownroute with

higherDSN?

SendRREP

yes

no no

sets up a reversepathpointing to

the source

TTL-- thenre-broadcast

RREQ

RREQalreadyseen?

no

DiscardRREQ

yes

store RREQ

myAddr=target?

87

L

H

J

D

C

G

S

N

K

BA

Route discovery succeeded !

Route Discovery

89

L

H

JC

G

D

S

N

K

B

A

RREP

Route Reply

Place route record andcost into RREP

(cost depends ontarget QoS metric).

DSN is increased

RREP sent onreverse path

myAddr=target?

yes

Nodes on the route set a forward pathto D as they receive the RREP


27/37

27

90

L

H

J

D

C

G

S

N

K

B

A

DATA

Data Delivery

Send data packet tonext hop using routing

table entry(no route in packet

header)

91

Remarks

D replies to the first arrived RREQ thusAODV favors the least congested route(that is not always the shortest one)

As OLSR, it only supports symmetric links!RREP forwarded along the route followed bythe RREQ

92

Timeouts

A routing table entry maintaining areverse path is purged after a timeoutinterval long enough to let the RREPcome back to the source

A routing table entry is purged if theroute has not been used for anactive_route_timeoutinterval


28/37

28

93

Route Maintenance

HELLO messages are periodically exchangedamong neighboring nodes to maintain localconnectivity

Absence of Hello messages is used as anindication of link failure

Hello messages can also contain the ID of theneighbors from which a node has heard

A node can also use MAC acknowledgementsor packet retransmissions done by a neighborto check on the next hop reachability

95

Link Failure

S

BC

F

D

G

E

Route Table (C)D: EDSN: 5Active ...

RERR

RERR

C sends RERRto upstreamneighbors

yes

C detectslink failure

C increases DSNand includes it in

the RERR

Route Table (C)D: EDSN: 6Inactive ...

RREQ

RREQ

Route Table (S)D: BDSN: 5Active ...

Route Table (S)D: BDSN: 6Inactive ...

RREQ

RREQ

RREQ

Route DSN:5Route DSN:6Route DSN:7

96

Link Failure: A Remark

Loop formation avoided thanks toDSN !

S B C

DE

F RREQ

RERR

RREQ

RREQ

C-F-E-S-B-C


29/37

29

97

Advantages

No need to include route in the packet headerNodes maintain routing tables containing entries onlyfor routes in use

DSNs allow nodes to avoid invalid/broken routesDSNs prevent formation of loops

Disadvantages

" Unused routes expire even if topology is unchanged" Multiple routes for each pair S-D are not supported" Need for symmetrical links

Advantages & Disadvantages

98

Routing Performance

Several performance comparisons[Broch98Mobicom,Johansson99Mobicom,Das00Infocom,Important03Infocom]

99

QoS Routing

Associate each link with a cost E.g., transmission power, transmitters residual

energy, traffic load, available bandwidth

RREQs used to collect the cost (or minimumcost) of all traversed links along a route

S

BC

F

D

G

E

S-B TxPw

RREQ

+ B-C TxPw + C-E TxPw + E-D TxPw

+ F-G TxPw+ G-D TxPw+ C-F TxPw


30/37

30

100

QoS Routing

Destination responds with a Route Reply to aRREQ if

it is the first RREQ with a given (current)sequence number, or

the route cost is smaller than all otherroutes cost received with the currentsequence number

Possible modifications to routing protocols,such as DSR, to make them QoS-aware

101

QoS Routing

Optimal routing decision requires constantupdates on link state information (e.g., delay,bandwidth, loss rate, link stability) resultingin large overhead

Dynamic nature of ad hoc networks makesmaintaining the precise link state information

very hard Even after resource reservation, QoS still

cannot be guaranteed due to the frequenttopology changes

Routing in VANETs


31/37

31

Which Protocol Is Suitable?

Proactive and reactive protocols may not be suitable: Proactive protocols: routing tables might often

contain stale information, due to the highly-dynamicnetwork topology

Reactive protocols typically imply high latency dueto the route discovery phase

Positioning systems can be exploited to efficientlyroute traffic in VANETs geographic routing

Geographic Routing (1)

Position-based schemes: nodes determine theirgeographical position through GPS; using Hello

messages a node can also acquire the position of its

neighboring nodes

It is assumed that the destination coordinates areknown (easy if its an RSU) and they are included in the

data packet

If the destination is a vehicle, then its position canvary over time and message delivery may fail

Geographic Routing (2)

Next hops are determined based on their position ormovement direction with respect to the intendeddestination

Self-election is possible, or the current forwardercan select the next one

In the basic greedy geographic technique, the packet isforwarded at each step to the neighbor whose position isclosest to destination

Greedy forwarding, however, is susceptible to problemof dead-ends


32/37

32

106

Dead-end with Geographic Routing

Both Es neighbors (i.e., C and D) arefurther away from the destination thanthe sender (E)

Adopt the right-hand rule to solve theproblem: E-C-F-B (same as to get out ofa maze)

A similar approach is used in sensornetworks; however in VANETs, several

factors can help:

the spatial-constrained topology(roads)

the delay-tolerant nature of thetraffic, along with the fact that

nodes move and can get closer to

the destination themselves

Mobility Models

109

Taxonomy of Mobility Models


33/37

33

110

Random Walk

Equal probability ofhopping in anydirection

111

Random Waypoint[Johnson and Maltz96]

Node at Xn: Picks next waypoint Xn+1 uniformly in area Picks speed Vn uniformly in [vmin,vmax] Moves to Xn+1 with speed Vn Pause atXn+1 for Tpause

Xn

Xn+1

112

Problems with RWP (1)

Distributions of node speed, position,distances, etc change with time

Node speed:100 users average

1 user

Time (s)

Speed(m/s)


34/37

34

113

Problems with RWP (2)

Distributions of node speed, position,distances, etc change with time

Distribution of node position:

Time = 0 sec Time = 2000 sec

114

Non-uniform Node Distribution (3)

115

Why does it matter ?

A.In the mobile case, thenodes are more oftentowards the center,distance between nodes isshorter, performance isbetter

The comparison is flawed.Should we use for staticcase the same distributionof node location as randomwaypoint. Is there such adistribution to compareagainst?

Random waypoint

Static

A (true) example: Compare impactof mobility on a protocol:

Experimenter places nodesuniformly for static case,according to random waypointfor mobile case

Finds that static is better Q.Find the bug !


35/37

35

116

Issues with Mobility Models

Is there a stable distribution of thesimulation state (time-stationarydistribution), reached if we run thesimulation long enough ?

If so: How long is long enough ? If it is too long, is there a way to get to

the stable distribution without running longsimulations (perfect simulation)?

117

Random Direction[Nain05]

Select independently fromone another:

1. (uniform) direction and speed2. travel time At borders: wrap-around or

reflection

118

Random Direction Properties

If nodes are uniformly distributed in thearea at t=0

Then, for any t>0, the nodes will beuniformly distributed in the area, providedthey move independently of each other


36/37

36

129

References on MANETs & Routing

Tutorial by Nitin Vaidya: http://www.crhc.uiuc.edu/~nhv/

IETF MANET Working Group http://www.ietf.org/html.charters/

manet-charter.html

http://protean.itd.nrl.navy.mil/manet/manet_home.html

130

References on Routing

C.E. Perkins, P. Bhagwat, Highly dynamic DestinationSequenced Distance Vector routing (DSDV) for mobilecomputers, Sigcomm 1994

Z.J. Haas, M.R. Pearlman, The Zone Routing Protocol(ZRP) for ad hoc networks, , 1999

B. Karp, H. T. Kung. GPSR: Greedy perimeter statelessrouting for wireless networks, MobiCom 2000

S. Basagni, I. Chlamtac, V.R. Syrotiuk, B.A. Woodward,A distance routing effect algorithm for mobility(DREAM), MobiCom1998

N. Guba, T. Camp, GLS: a location service for an ad hocnetwork,Grace Hopper Celebration2002

131


[Johnson96] D.B. Johnson, D.A. Maltz, "Dynamicsource routing in ad hoc wireless networks", MobileComputing, vol. 353, 1996

[Perkins99] C.E. Perkins, E.M. Royer, "Ad hoc on-demand distance vector routing," 2nd IEEE Workshopon Mobile Computing Systems and Applications, 1999

J. Broch, D.A. Maltz, D.B. Johnson et al., Aperformance comparison of multi-hop wireless ad hocnetwork routing protocols, MobiCom 1998

AODV Linux freeware implementations: http://moment.cs.ucsb.edu/AODV/aodv.html#Implementations


37/37

132

S.R. Das, C.E. Perkins, E.M. Royer, PerformanceComparison of two on-demand routing protocols for adhoc networks, Infocom 2000

P. Johansson, T. Larsson, N. Hedman et al., Scenario-based performance analysis of routing protocols formobile ad-hoc networks, MobiCom 1999

F. Bai, N. Sadagopan, A. Helmy, " IMPORTANT: Aframework to systematically analyze the Impact ofMobility on Performance of RouTing protocols for Adhoc NeTworks," Infocom2003

Two papers by A. Helmy et al. at ICC 2004


Documents

Routing in Wireless Ad Hoc Networks