[PDF] - A Review of Curren t Routing Proto cols for Ad-Ho c Mobile PDF Document

SLIDE 1 A Review

f

Curren t Routing Proto cols for Ad-Ho c Mobile Wireless Net w

rks

Elizab eth M. Ro y er Dept.

f

Electrical & Computer Engineering Univ ersit y

f

California, San ta Barbara er

yer@alpha.e

c e.ucsb.e du C-K T

h

Dept.

f

Electrical & Computer Engineering Georgia Institute

f

T ec hnology , A tlan ta cktoh@e e.gate ch.e du Abstract An ad-ho c mobile net w

rk

is a collection

f

mobile no des that are dynamically and arbitrarily lo cated in suc h a manner that the in terconnections b et w een no des are capable

f

c hanging

n

a con tin ual basis. In

rder

to facilitate comm unicatio n within the net w

rk,

a routing proto col is used to disco v er routes b et w een no des. The primary goal

f

suc h an ad-ho c net w

rk

routing proto col is correct and ecien t route establishmen t b et w een a pair

f

no des so that messages ma y b e deliv ered in a timely manner. Route construction should b e done with a minim um

f
v

erhead and bandwidth consumption. This pap er examines routing proto cols for ad-ho c net w

rks

and ev aluates these proto cols based

n

a giv en set

f

parameters. The pap er pro vides an

v

erview

f

eigh t dieren t proto cols b y presen ting their c haracteristics and functionalit y , and then pro vides a comparison and discussion

f

their resp ectiv e merits and dra wbac ks. 1 In tro duction Since their emergence in the 1970s, wireless net w

rks

ha v e b ecome increasingly p

pular

in the computing industry . This is particularly true within the past decade whic h has seen wireless net w

rks

b eing adapted to enable mobilit y . There are curren tly t w

v

ariations

f

mobile wireless net w

rks.

The rst is kno wn as infrastructured net w

rks,

i.e., those net w

rks

with xed and wired gatew a ys. The bridges for these net w

rks

are kno wn as base stations. A mobile unit within these net w

rks

con- nects to, and comm unicates with, the nearest base station that is within its comm unication radius. As the mobile tra v els

ut
f

range

f
ne

base station and in to the range

f

another, a \hando "

ccurs

from the

ld

base station to the new, and the mobile is able to con tin ue comm unication seam- lessly throughout the net w

rk.

T ypical applications

f

this t yp e

f

net w

rk

include

ce

wireless lo cal area net w

rks

(WLANs). The second t yp e

f

mobile wireless net w

rk

is the infrastructureless mobile net w

rk,

commonly kno wn as an ad-ho c net w

rk.

Infrastructureless net w

rks

ha v e no xed routers; all no des are capable

f

mo v emen t and can b e connected dynamically in an arbitrary manner. No des

f

these net w

rks

function as routers whic h disco v er and main tain routes to

ther

no des in the net w

rk.

Example applications

f

ad-ho c net w

rks

are emergency searc h-and-rescue

p

erations, meetings

r

con v en tions in whic h p ersons wish to quic kly share information, and data acquisition

p

erations in inhospitable terrains. This pap er examines routing proto cols designed for these ad-ho c net w

rks

b y rst describing the

p

eration

f

eac h

f

the proto cols and then comparing their v arious c haracteristics. The remainder

f

the pap er is

rganized

as follo ws: Section 2 presen ts a discussion

f

t w

sub

divisions

f

ad-ho c routing proto cols. Subsection 2.1 discusses curren t table-driv en proto cols, while Subsection 2.2 describ es 1

SLIDE 2

Ad-Hoc Routing Protocols Table Driven DSDV CGSR WRP Source-initiated On-Demand Driven AODV DSR LMR ABR TORA SSR

Figure 1: Categorization

f

Ad-Ho c Routing Proto cols. those proto cols whic h are classied as

n-demand.

Section 3 presen ts qualitativ e comparisons

f

table-driv en proto cols, follo w ed b y

n-demand-driv

en proto cols, and nally a general comparison

f

table-driv en and

n-demand

proto cols. Applications and c hallenges facing ad-ho c mobile wireless net w

rks

are discussed in Section 4. Finally , Section 5 concludes the pap er. 2 Existing Ad-Ho c Routing Proto cols Since the adv en t

f

D ARP A pac k et radio net w

rks

in the early 1970s [11 ], n umerous proto cols ha v e b een dev elop ed for ad-ho c mobile net w

rks.

Suc h proto cols m ust deal with the t ypical limitations

f

these net w

rks,

whic h include high p

w

er consumption, lo w bandwidth, and high error rates. As sho wn in Figure 1, these routing proto cols ma y generally b e categorized as: (a) table-driv en and (b) source-initiated

n-demand

driv en. Solid lines in this gure represen t direct descendan ts while dotted lines depict logical descendan ts. Despite b eing designed for the same t yp e

f

underlying net w

rk,

the c haracteristics

f

eac h

f

these proto cols are quite distinct. The follo wing sections describ e the proto cols and categorize them according to their c haracteristics. 2.1 T able-Driv en Routing Proto cols The table-driv en routing proto cols attempt to main tain consisten t, up-to-date routing information from eac h no de to ev ery

ther

no de in the net w

rk.

These proto cols require eac h no de to main tain

ne
r

more tables to store routing information, and they resp

nd

to c hanges in net w

rk

top

logy

b y propagating up dates throughout the net w

rk

in

rder

to main tain a consisten t net w

rk

view. The areas where they dier are the n um b er

f

necessary routing-related tables and the metho ds b y whic h c hanges in net w

rk

structure are broadcast. The follo wing sections discuss some

f

the existing table-driv en ad-ho c routing proto cols. 2.1.1 Destination-Sequenced Distance-V ector Routing (DSD V) The Destination-Sequenced Distance-V ector Routing proto col (DSD V) describ ed in [17 ] is a table- driv en algorithm based

n

the classical Bellman-F

rd

routing mec hanism [7 ]. The impro v emen ts made to the Bellman-F

rd

algorithm include freedom from lo

ps

in routing tables. Ev ery mobile no de in the net w

rk

main tains a routing table in whic h all

f

the p

ssible

desti- nations within the net w

rk

and the n um b er

f

hops to eac h destination are recorded. Eac h en try is mark ed with a sequence n um b er assigned b y the destination no de. The sequence n um b ers enable the 2

SLIDE 3 mobile no des to distinguish stale routes from new

nes,

thereb y a v

iding

the formation

f

routing lo

ps.

Routing table up dates are p erio dically transmitted throughout the net w

rk

in

rder

to main- tain table consistency . T

help

alleviate the p

ten

tially large amoun t

f

net w

rk

trac that suc h up dates can generate, route up dates can emplo y t w

p
ssible

t yp es

f

pac k ets. The rst is kno wn as a \full dump." This t yp e

f

pac k et carries all a v ailable routing information and can require m ultiple net w

rk

proto col data units (NPDUs). During p erio ds

f
ccasional

mo v emen t, these pac k ets are transmitted infrequen tly . Smaller \incremen tal" pac k ets are used to rela y

nly

that information whic h has c hanged since the last full dump. Eac h

f

these broadcasts should t in to a standard size NPDU, thereb y decreasing the amoun t

f

trac generated. The mobile no des main tain an additional table where they store the data sen t in the incremen tal routing information pac k ets. New route broadcasts con tain the address

f

the destination, the n um b er

f

hops to reac h the destination, the sequence n um b er

f

the information receiv ed regarding the destination, as w ell as a new sequence n um b er unique to the broadcast [17 ]. The route lab eled with the most recen t sequence n um b er is alw a ys used. In the ev en t that t w

up

dates ha v e the same sequence n um b er, the route with the smaller metric is used in

rder

to

ptimize

(shorten) the path. Mobiles also k eep trac k

f

the settling time

f

routes,

r

the w eigh ted a v erage time that routes to a destination will uctuate b efore the route with the b est metric is receiv ed (see [17 ]). By dela ying the broadcast

f

a routing up date b y the length

f

the settling time, mobiles can reduce net w

rk

trac and

ptimize

routes b y eliminating those broadcasts that w

uld
ccur

if a b etter route w as disco v ered in the v ery near future. 2.1.2 Clusterhead Gatew a y Switc h Routing (CGSR) The Clusterhead Gatew a y Switc h Routing (CGSR) proto col diers from the previous proto col in the t yp e

f

addressing and net w

rk
rganization

sc heme emplo y ed. Instead

f

a \at" net w

rk,

CGSR is a clustered m ultihop mobile wireless net w

rk

with sev eral heuristic routing sc hemes [4]. The authors state that b y ha ving a cluster head con trolling a group

f

ad-ho c no des, a framew

rk

for co de separation (among clusters), c hannel access, routing and bandwidth allo cation can b e ac hiev ed. A cluster head selection algorithm is utilized to elect a no de as the cluster head using a distributed algorithm within the cluster. The disadv an tage

f

ha ving a cluster head sc heme is that frequen t cluster head c hanges can adv ersely aect routing proto col p erformance since no des are busy in cluster head selection rather than pac k et rela ying. Hence, instead

f

in v

king

cluster head reselection ev ery time the cluster mem b ership c hanges, a Least Cluster Change (LCC) clustering algorithm is in tro duced. Using LCC, cluster heads

nly

c hange when t w

cluster

heads come in to con tact,

r

when a no de mo v es

ut
f

con tact

f

all

ther

cluster heads.

1 2 3 4 5 6 7 8

Cluster head Gateway Node

Figure 2: CGSR: Routing from No de 1 to No de 8. CGSR uses DSD V as the underlying routing sc heme, and hence has m uc h

f

the same

v

erhead as DSD V. Ho w ev er, it mo dies DSD V b y using a hierarc hical cluster head-to-gatew a y routing approac h 3

SLIDE 4 to route trac from source to destination. Gatew a y no des are no des that are within comm unication range

f

t w

r

more cluster heads. A pac k et sen t b y a no de is rst routed to its cluster head, and then the pac k et is routed from the cluster head to a gatew a y to another cluster head, and so

n

un til the cluster head

f

the destination no de is reac hed. The pac k et is then transmitted to the destination. Figure 2 illustrates an example

f

this routing sc heme. Using this metho d, eac h no de m ust k eep a \cluster mem b er table" where it stores the destination cluster head for eac h mobile no de in the net w

rk.

These cluster mem b er tables are broadcast b y eac h no de p erio dically using the DSD V algorithm. No des up date their cluster mem b er tables

n

the reception

f

suc h a table from a neigh b

r.

In addition to the cluster mem b er table, eac h no de m ust also main tain a routing table, whic h is used to determine the next hop in

rder

to reac h the destination. On receiving a pac k et, a no de will consult its cluster mem b er table and routing table to determine the nearest cluster head along the route to the destination. Next the no de will c hec k its routing table to determine the no de in

rder

to reac h the selected cluster head. It then transmits the pac k et to this no de. 2.1.3 The Wireless Routing Proto col (WRP) The Wireless Routing Proto col (WRP) describ ed in [14 ] is a table-based proto col with the goal

f

main taining routing information among all no des in the net w

rk.

Eac h no de in the net w

rk

is resp

nsible

for main taining four tables: (a) distance table, (b) routing table, (c) link-cost table, and (d) message retransmission list (MRL) table. Eac h en try

f

the MRL con tains the sequence n um b er

f

the up date message, a retransmission coun ter, an ac kno wledgmen t-required ag v ector with

ne

en try p er neigh b

r,

and a list

f

up dates sen t in the up date message. The MRL records whic h up dates in an up date message need to b e retransmitted and whic h neigh b

rs

should ac kno wledge the retransmission [14 ]. Mobiles inform eac h

ther
f

link c hanges through the use

f

up date messages. An up date message is sen t

nly

b et w een neigh b

ring

no des and con tains a list

f

up dates (the destination, the distance to the destination, and the predecessor

f

the destination), as w ell as a list

f

resp

nses

indicating whic h mobiles should ac kno wledge (A CK) the up date. Mobiles send up date messages after pro cessing up dates from neigh b

rs
r

detecting a c hange in a link to a neigh b

r.

In the ev en t

f

the loss

f

a link b et w een t w

no

des, the no des send up date messages to their neigh b

rs.

The neigh b

rs

then up date their distance table en tries and c hec k for new p

ssible

paths through

ther

no des. An y new paths are rela y ed bac k to the

riginal

no des so that they can up date their tables accordingly . No des learn

f

the existence

f

their neigh b

rs

from the receipt

f

ac kno wledgmen ts and

ther

messages. If a no de is not sending messages, it m ust send a hel lo message within a sp ecied time p erio d to ensure connectivit y . Otherwise, the lac k

f

messages from the no de indicates the failure

f

that link; this ma y cause a false alarm. When a mobile receiv es a hel lo message from a new no de, that new no de is added to the mobile's routing table, and the mobile sends the new no de a cop y

f

its routing table information. P art

f

the no v elt y

f

WRP stems from the w a y in whic h it ac hiev es lo

p

freedom. In WRP , routing no des comm unicate the distance and second-to-last hop information for eac h destination in the wireless net w

rks.

WRP b elongs to the class

f

path nding algorithms with an imp

rtan

t exception. It a v

ids

the \coun t-to-innit y" problem [21 ] b y forcing eac h no de to p erform consistency c hec ks

f

predecessor information rep

rted

b y all its neigh b

rs.

This ultimately (though not instan taneously) eliminates lo

ping

situations and pro vides faster route con v ergence when a link failure ev en t

ccurs.

4

SLIDE 5

N1 N2 N3 N4 N5 N6 N7 N8 N1 N2 N3 N4 N5 N6 N7 N8 N1 N2 N3 N4 N5 N6 N7 N8 N1 N2 N3 N4 N5 N6 N7 N8 N1 N2 N3 N4 N5 N6 N7 N8 N1 N2 N3 N4 N5 N6 N7 N8

(a) Propagation of the RREQ (b) Path of the RREP to the Source Destination Destination Source Source

Figure 3: A OD V Route Disco v ery . 2.2 Source-Initiated On-Demand Routing A dieren t approac h from table-driv en routing is source-initiated

n-demand

routing. This t yp e

f

routing creates routes

nly

when desired b y the source no de. When a no de requires a route to a destination, it initiates a route disco v ery pro cess within the net w

rk.

This pro cess is completed

nce

a route is found

r

all p

ssible

route p erm utations ha v e b een examined. Once a route has b een established, it is main tained b y some form

f

route main tenance pro cedure un til either the destination b ecomes inaccessible along ev ery path from the source

r

un til the route is no longer desired. 2.2.1 Ad-ho c On-Demand Distance V ector Routing (A OD V) The Ad-ho c On-Demand Distance V ector (A OD V) routing proto col describ ed in [19 ] builds

n

the DSD V algorithm previously describ ed. A OD V is an impro v emen t

n

DSD V b ecause it t ypically minimizes the n um b er

f

required broadcasts b y creating routes

n

an

n-demand

basis, as

pp
sed

to main taining a complete list

f

routes as in the DSD V algorithm. The authors

f

A OD V classify it as a pur e

n-demand

r

ute

ac quisition system, as no des that are not

n

a selected path do not main tain routing information

r

participate in routing table exc hanges [19 ]. When a source no de desires to send a message to some destination no de and do es not already ha v e a v alid route to that destination, it initiates a Path Disc

very

pro cess to lo cate the

ther

no de. It broadcasts a route request (RREQ) pac k et to its neigh b

rs,

whic h then forw ard the request to their neigh b

rs,

and so

n,

un til either the destination

r

an in termediate no de with a \fresh enough" route to the destination is lo cated. Figure 3a illustrates the propagation

f

the broadcast RREQs across the net w

rk.

A OD V utilizes destination sequence n um b ers to ensure all routes are lo

p-free

and con tain the most recen t route information. Eac h no de main tains its

wn

sequence n um b er, as w ell as a broadcast ID. The broadcast ID is incremen ted for ev ery RREQ the no de initiates, and together with the no de's IP address, uniquely iden ties a RREQ. Along with its

wn

sequence n um b er and the broadcast ID, the source no de includes in the RREQ the most recen t sequence n um b er it has for the destination. In termediate no des can reply to the RREQ

nly

if they ha v e a route to the destination whose corresp

nding

destination sequence n um b er is greater than

r

equal to that con tained in the RREQ. During the pro cess

f

forw arding the RREQ, in termediate no des record in their route tables the address

f

the neigh b

r

from whic h the rst cop y

f

the broadcast pac k et is receiv ed, thereb y establishing a rev erse path. If additional copies

f

the same RREQ are later receiv ed, these pac k ets are discarded. Once the RREQ reac hes the destination

r

an in termediate no de with a fresh enough route, the destination/in termediate no de resp

nds

b y unicasting a route reply (RREP) pac k et bac k to the neigh b

r

from whic h it rst receiv ed the RREQ (Figure 3b). As the RREP is routed bac k 5

SLIDE 6

(b) Propagation of the Route Reply with the Route Record

N2

N3 N4 N5

N6 N7 N8

N1

Destination Source

N1-N2-N5-N8 N1-N2-N5-N8 N1-N2-N5-N8

N2

N3 N4 N5

N6 N7 N8

N1

Destination Source

N1 N1-N2 N1 N1-N3 N1-N2-N5 N1-N3-N4 N1-N3-N4 N1-N3-N4 N1-N3-N4-N6 N1-N3-N4-N7

(a) Building of the Route Record during Route Discovery

Figure 4: Creation

f

the R

ute

R e c

r

d in DSR. along the rev erse path, no des along this path set up forw ard route en tries in their route tables whic h p

in

t to the no de from whic h the RREP came. These forw ard route en tries indicate the activ e forw ard route. Asso ciated with eac h route en try is a route timer whic h will cause the deletion

f

the en try if it is not used within the sp ecied lifetime. Because the RREP is forw arded along the path established b y the RREQ, A OD V

nly

supp

rts

the use

f

symmetric links. Routes are main tained as follo ws. If a source no de mo v es, it is able to reinitiate the route disco v ery proto col to nd a new route to the destination. If a no de along the route mo v es, its upstream neigh b

r

notices the mo v e and propagates a link failur e notic ation message (a RREP with innite metric) to eac h

f

its activ e upstream neigh b

rs

to inform them

f

the erasure

f

that part

f

the route [19 ]. These no des in turn propagate the link failur e notic ation to their upstream neigh b

rs,

and so

n

un til the source no de is reac hed. The source no de ma y then c ho

se

to re-initiate route disco v ery for that destination if a route is still desired. An additional asp ect

f

the proto col is the use

f

hel lo messages, p erio dic lo cal broadcasts b y a no de to inform eac h mobile no de

f
ther

no des in its neigh b

rho
d.

Hello messages can b e used to main tain the lo cal connectivit y

f

a no de. Ho w ev er the use

f

hello messages is not required. No des listen for retransmissions

f

data pac k ets to ensure the next hop is still within reac h. If suc h a retransmission is not heard, the no de ma y use an y

ne
f

a n um b er

f

tec hniques, including the reception

f

hello messages, to determine whether the next hop is within comm unication range. The hel lo messages ma y list the

ther

no des from whic h a mobile has heard, thereb y yielding a greater kno wledge

f

the net w

rk

connectivit y . 2.2.2 Dynamic Source Routing (DSR) The Dynamic Source Routing (DSR) proto col presen ted in [10 ] is an

n-demand

routing proto col that is based

n

the concept

f

source routing. Mobile no des are required to main tain route cac hes that con tain the source routes

f

whic h the mobile is a w are. En tries in the route cac he are con tin ually up dated as new routes are learned. The proto col consists

f

t w

ma

jor phases: route disco v ery and route main tenance. When a mobile no de has a pac k et to send to some destination, it rst consults its route cac he to determine whether it already has a route to the destination. If it has an unexpired route to the destination, it will use this route to send the pac k et. On the

ther

hand, if the no de do es not ha v e suc h a route, it initiates route disco v ery b y broadcasting a r

ute

r e quest pac k et. This r

ute

r e quest con tains the address

f

the destination, along with the source no de's address and a unique iden tication n um b er. Eac h no de receiving the pac k et c hec ks whether it kno ws

f

a route to the destination. If it do es not, it adds its

wn

address to the r

ute

r e c

r

d

f

the pac k et and then forw ards the pac k et along its

utgoing

links. T

limit

the n um b er

f

r

ute

r e quests propagated

n

the

utgoing

links

f

a no de, a 6

SLIDE 7

Destination Source

Ad Hoc Node ’height’ metric

(a)

(1) (2) (3) (4) A B C G F E D A B C D E F G A B C D E F G A B C D E F G Link Reversal Link Failure

(b)

Figure 5: (a) Route creation (sho wing link direction assignmen t), and(b) Route Main tenace (sho wing link rev ersal phenonemon) in TORA. mobile

nly

forw ards the r

ute

r e quest if the request has not y et b een seen b y the mobile and if the mobile's address do es not already app ear in the r

ute

r e c

r

d. A r

ute

r eply is generated when either the r

ute

r e quest reac hes the destination itself,

r

when it reac hes an in termediate no de whic h con tains in its route cac he an unexpired route to the destination [2 ]. By the time the pac k et reac hes either the destination

r

suc h an in termediate no de, it con tains a r

ute

r e c

r

d yielding the sequence

f

hops tak en. Figure 4a illustrates the formation

f

the r

ute

r e c

r

d as the r

ute

r e quest propagates through the net w

rk.

If the no de generating the r

ute

r eply is the destination, it places the r

ute

r e c

r

d con tained in the r

ute

r e quest in to the r

ute

r eply. If the resp

nding

no de is an in termediate no de, it will app end its cac hed route to the r

ute

r e c

r

d and then generate the r

ute

r eply. T

return

the r

ute

r eply, the resp

nding

no de m ust ha v e a route to the initiator. If it has a route to the initiator in its route cac he, it ma y use that route. Otherwise, if symmetric links are supp

rted,

the no de ma y rev erse the route in the r

ute

r e c

r

d. If symmetric links are not supp

rted,

the no de ma y initiate its

wn

route disco v ery and piggybac k the r

ute

r eply

n

the new r

ute

r e quest. Figure 4b sho ws the transmission

f

the r

ute

r eply with its asso ciated r

ute

r e c

r

d bac k to the source no de. Route main tenance is accomplished through the use

f

r

ute

err

r

pac k ets and ac kno wledgmen ts. R

ute

err

r

pac k ets are generated at a no de when the data link la y er encoun ters a fatal transmission problem. When a r

ute

err

r

pac k et is receiv ed, the hop in error is remo v ed from the no de's route cac he and all routes con taining the hop are truncated at that p

in

t. In addition to r

ute

err

r

messages, ac kno wledgmen ts are used to v erify the correct

p

eration

f

the route links. Suc h ac kno wledgmen ts include passiv e ac kno wledgmen ts, where a mobile is able to hear the next hop forw arding the pac k et along the route. 2.2.3 T emp

rally-Ordered

Routing Algorithm (TORA) TORA (T emp

rally-Ordered

Routing Algorithm) is a highly adaptiv e, lo

p-free,

distributed routing algorithm based

n

the concept

f

link rev ersal [16 ]. TORA is prop

sed

to

p

erate in a highly dynamic mobile net w

rking

en vironmen t. It is source-initiated and pro vides m ultiple routes for an y desired source/destination pair. The k ey design concept

f

TORA is the lo calization

f

con trol mes- 7

SLIDE 8 sages to a v ery small set

f

no des near the

ccurrence
f

a toplogical c hange. T

accomplish

this, no des need to main tain routing information ab

ut

adjacen t (1-hop) no des. The proto col p erforms three basic functions: (a) route creation, (b) route main tainence, and (c) route erasure. During the route creation and main tenance phases, no des use a \heigh t" metric to establish a directed acyclic graph (D A G) ro

ted

at the destination. Thereafter, links are assigned a direction (upstream

r

do wnstream) based

n

the relativ e heigh t metric

f

neigh b

ring

no des, as sho wn in Figure 5a. This pro cess

f

establishing a D A G is similar to the query/reply pro cess prop

sed

in LMR (Ligh t w eigh t Mobile Routing) [5 ]. In times

f

no de mobilit y , the D A G route is brok en and route main tenace is necessary to re-establish a D A G ro

ted

at the same destination. As sho wn in Figure 5b, up

n

failure

f

the last do wnstream link, a no de generates a new reference lev el whic h results in the propagation

f

that reference lev el b y neigh b

ring

no des, eectiv ely co

rdinating

a structured reaction to the failure. Links are rev ersed to reect the c hange in adapting to the new reference lev el. This has the same eect as rev ersing the direction

f
ne
r

more links when a no de has no do wnstream links. Timing is an imp

rtan

t factor for TORA b ecause the \heigh t" metric is dep enden t

n

the logical time

f

a link failure; TORA assumes all no des ha v e sync hronized clo c ks (accomplished via an external time source suc h as Global P

sitioning

System). TORA's metric is a quin tuple comprised

f

v e elemen ts, namely: (a) logical time

f

a link failure, (b) the unique ID

f

the no de that dened the new reference lev el, (c) a reection indicator bit, (d) a propagation

rdering

parameter, and (e) the unique ID

f

the no de. The rst three elemen ts collectiv ely represen t the reference lev el. A new reference lev el is dened eac h time a no des loses its last do wnstream link due to a link failure. TORA's route erasure phase essen tially in v

lv

es

ding

a broadcast \clear pac k et" (CLR) throughout the net w

rk

to erase in v alid routes. In TORA, there is a p

ten

tial for

scillations

to

ccur,

esp ecially when m ultiple sets

f

co

rdi-

nating no des are concurren tly detecting partitions, erasing routes, and building new routes based

n

eac h

ther.

Because TORA uses in terno dal co

rdination,

its instabilit y problem is similar to the \coun t-to-innit y" problem in distance-v ector routing proto cols, except that suc h

scillations

are temp

rary

and route con v ergence will ultimately

ccur.

2.2.4 Asso ciativit y-Based Routing (ABR) A totally dieren t approac h in mobile routing is prop

sed

in [22 ]. The Asso ciativit y-Based Routing (ABR) proto col is free from lo

ps,

deadlo c k, and pac k et duplicates, and denes a new routing metric for ad-ho c mobile net w

rks.

This metric is kno wn as the de gr e e

f

asso ciation stability. In ABR, a route is selected based

n

the degree

f

asso ciation stabilit y

f

mobile no des. Eac h no de p erio dically generates a b eacon to signify its existence. When receiv ed b y neigh b

ring

no des, this b eaconing causes their asso ciativit y tables to b e up dated. F

r

eac h b eacon receiv ed, the asso ciativit y tic k

f

the curren t no de with resp ect to the b eaconing no de is incremen ted. Asso ciation stabilit y is dened b y connection stabilit y

f
ne

no de with resp ect to another no de

v

er time and space. A high degree

f

asso ciation stabilit y ma y indicate a lo w state

f

no de mobilit y , while a lo w degree ma y indicate a high state

f

no de mobilit y . Asso ciativit y tic ks are reset when the neigh b

rs
f

a no de

r

the no de itself mo v es

ut
f

pro ximit y . A fundamen tal

b

jectiv e

f

ABR is to deriv e longer-liv ed routes for ad-ho c mobile net w

rks.

The three phases

f

ABR are: (a) route disco v ery , (b) route re-construction (RR C), and (c) route deletion. The route disco v ery phase is accomplished b y a broadcast query and a w ait-reply (BQ-REPL Y) cycle. A no de desiring a route broadcasts a BQ message in searc h

f

mobiles that ha v e a route to the destination. All no des receiving the query (that are not the destination) app end their addresses and their asso ciativit y tic ks with their neigh b

rs

along with QoS information to the 8

SLIDE 9

(b) Route Maintenance for a Destination Move SRC DEST’ DEST LQ[H] RN[0] RN[0] H=3 SRC SRC’ RN[1] BQ DEST (a) Route Maintenance for a Source Move

Figure 6: Route Main tenance for Source and Destination Mo v emen t in ABR. query pac k et. A successor no de erases its upstream no de neigh b

rs'

asso ciativit y tic k en tries and retains

nly

the en try concerned with itself and its upstream no de. In this w a y , eac h resultan t pac k et arriving at the destination will con tain the asso ciativit y tic ks

f

the no des along the route to the destination. The destination is then able to select the b est route b y examining the asso ciativit y tic ks along eac h

f

the paths. In the case where m ultiple paths ha v e the same

v

erall degree

f

asso ciation stabilit y , the route with the minim um n um b er

f

hops is selected. The destination then sends a REPL Y pac k et bac k to the source along this path. No des propagating the REPL Y mark their routes as v alid. All

ther

routes remain inactiv e and the p

ssibilit

y

f

duplicate pac k ets arriving at the destination is a v

ided.

Route re-construction ma y consist

f

partial route disco v ery , in v alid route erasure, v alid route up dates, and new route disco v ery , dep ending

n

whic h no de(s) along the route mo v e. Mo v emen t b y the source results in a new BQ-REPL Y pro cess, as sho wn in Figure 6a. The RN[1] message is a route notication that is used to erase the route en tries asso ciated with do wnstream no des. When the destination no de mo v es, the immediate upstream no de erases its route and determines if the no de is still reac hable b y a lo calized query (LQ[H]) pro cess, where H refers to the hop coun t from the upstream no de to the destination (Figure 6b). If the destination receiv es the LQ pac k et, it REPL Ys with the b est partial route;

therwise,

the initiating no de times

ut

and the pro cess bac ktrac ks to the next upstream no de. Here an RN[0] message is sen t to the next upstream no de to erase the in v alid routes and inform this no de it should in v

k

e the LQ[H] pro cess. If this pro cess results in bac ktrac king more than halfw a y to the source, the LQ pro cess is discon tin ued and a new BQ pro cess is initiated at the source. When a disco v ered route is no longer desired, the source no de initiates a route delete (RD) broadcast so that all no des along the route up date their routing tables. The RD message is propagated b y a full broadcast, as

pp
sed

to a directed broadcast, b ecause the source no de ma y not b e a w are

f

an y route no de c hanges that

ccurred

during route re-constructions. 2.2.5 Signal Stabilit y Routing (SSR) Another

n-demand

proto col is the Signal Stabilit y based Adaptiv e Routing proto col (SSR) presen ted in [6]. Unlik e the algorithms describ ed so far, SSR selects routes based

n

the signal strength b et w een no des and

n

a no de's lo cation stabilit y . This route selection criteria has the eect

f

c ho

sing

routes that ha v e \stronger" connectivities. SSR can b e divided in to t w

co
p

erativ e proto cols: the Dynamic Routing Proto col (DRP) and the Static Routing Proto col (SRP). The DRP is resp

nsible

for the main tenance

f

the Signal Stabilit y T able (SST) and the Routing T able (R T). The SST records the signal strength

f

neigh b

ring

no des, whic h is

btained

b y p erio dic b eacons from the link la y er

f

eac h neigh b

ring

no de. The signal strength ma y b e recorded as either a 9

SLIDE 10 strong

r

w eak c hannel. All transmissions are receiv ed b y , and pro cessed in, the DRP . After up dating all appropriate table en tries, the DRP passes a receiv ed pac k et to the SRP . The SRP pro cesses pac k ets b y passing the pac k et up the stac k if it is the in tended receiv er

r

lo

king

up the destination in the R T and then forw arding the pac k et if it is not. If no en try is found in the R T for the destination, a r

ute-se

ar ch pro cess is initiated to nd a route. Route requests are propagated throughout the net w

rk

but are

nly

forw arded to the next hop if they are receiv ed

v

er strong c hannels and ha v e not b een previously pro cessed (to prev en t lo

ping).

The destination c ho

ses

the rst arriving r

ute-se

ar ch pac k et to send bac k b ecause it is most probable that the pac k et arriv ed

v

er the shortest and/or least congested path. The DRP then rev erses the selected route and sends a r

ute-r

eply message bac k to the initiator. The DRP

f

the no des along the path up date their R Ts accordingly . R

ute-se

ar ch pac k ets arriving at the destination ha v e necessarily c hosen the path

f

strongest signal stabilit y , as the pac k ets are dropp ed at a no de if they ha v e arriv ed

v

er a w eak c hannel. If there is no r

ute-r

eply message receiv ed at the source within a sp ecic timeout p erio d, the source c hanges the PREF eld in the header to indicate that w eak c hannels are acceptable, as these ma y b e the

nly

links

v

er whic h the pac k et can b e propagated. When a failed link is detected within the net w

rk,

the in termediate no des send an error message to the source indicating whic h c hannel has failed. The source then initiates another r

ute-se

ar ch pro cess to nd a new path to the destination. The source also sends an erase message to notify all no des

f

the brok en link. 3 Comparisons The follo wing sections pro vide comparisons

f

the previously describ ed routing algorithms. Sec- tion 3.1 compares table-driv en proto cols, and Section 3.2 compares

n-demand

proto cols. Section 3.3 presen ts a discussion

f

the t w

classes
f

algorithms. In T ables 1 and 2, Time Complexit y is dened as the n um b er

f

steps needed to p erform a proto col

p

eration, and Comm unication Complexit y is the n um b er

f

messages needed to p erform a proto col

p

eration [5], [23 ]. Also, the v alues for these metrics represen t worst c ase b eha vior. 3.1 T able-Driv en Proto cols Our discussion here will b e based

n

T able 1. As stated earlier, DSD V routing is essen tially a mo dication

f

the basic Bellman-F

rd

routing algorithm. The mo dications include the guaran tee

f

lo

p-free

routes and a simple route up date proto col. While

nly

pro viding

ne

path to an y giv en destination, DSD V selects the shortest path based

n

the n um b er

f

hops to the destination. DSD V pro vides t w

t

yp es

f

up date messages,

ne
f

whic h is signican tly smaller than the

ther.

The smaller up date message can b e used for incremen tal up dates so that the en tire routing table need not b e transmitted for ev ery c hange in net w

rk

top

logy

. Ho w ev er, DSD V is inecien t b ecause

f

the requiremen t

f

p erio dic up date transmissions, regardless

f

the n um b er

f

c hanges in the net w

rk

top

logy

. This eectiv ely limits the n um b er

f

no des that can connect to the net w

rk

since the

v

erhead gro ws as O(n 2 ). In CGSR, DSD V is used as the underlying routing proto col. Routing in CGSR

ccurs
v

er cluster heads and gatew a ys. A cluster head table is necessary in addition to the routing table. One adv an tage

f

CGSR is that sev eral heuristic metho ds can b e emplo y ed to impro v e the proto col's p erformance. These metho ds include priorit y tok en sc heduling, gatew a y co de sc heduling, and path reserv ation [4 ]. 10

SLIDE 11 P arameter s DSD V CGSR WRP Time Complexit y (link addition / failure) O(d) O(d) O(h) Comm unicati

n

Complexit y (link addition / failure) O(x=N) O(x=N) O(x=N) Routing Philosoph y Flat Hierarc hical Flat 1 Lo

p

F ree Y es Y es Y es, but not instan taneou s Multicast Capabilit y No No 2 No Num b er

f

Required T ables Tw

Tw
F
ur

F requency

f

Up date T ransmissions P erio dically P erio dically P erio dically & as needed & as needed Up dates T ransmitted to Neigh b

rs

Neigh b

rs

Neigh b

rs

& cluster head Utilizes Sequence Num b ers Y es Y es Y es Utilizes \Hello" Messages Y es No Y es Critical No des No Y es (cluster head) No Routing Metric Shortest P ath Shortest P ath Shortest P ath T able 1: Comparisons

f

the Characteristics

f

T able-Driv en Routing Proto cols. Abbreviations: N = N umber

f

nodes in the netw

r

k d = N etw

r

k diameter h = H eig ht

f

r

uting

tr ee x = N umber

f

nodes af f ected by a topolog ical chang e The WRP proto col diers from the

ther

proto cols in sev eral w a ys. WRP requires eac h no de to main tain four routing tables. This can lead to substan tial memory requiremen ts, esp ecially when the n um b er

f

no des in the net w

rk

is large. F urthermore, the WRP proto col requires the use

f

hel lo pac k ets whenev er there are no recen t pac k et transmissions from a giv en no de. The hel lo pac k ets consume bandwidth and disallo w a no de to en ter sleep mo de. Ho w ev er, though it b elongs to the class

f

path nding algorithms, WRP has an adv an tage

v

er

ther

path nding algorithms b ecause it a v

ids

the problem

f

creating temp

rary

routing lo

ps

that these algorithms ha v e through the v erication

f

predecessor information, as describ ed in Section 2.1.3. Ha ving discussed the

p

eration and c haracteristics

f

eac h

f

the existing table-driv en based routing proto cols, it is imp

rtan

t to highligh t the dierences. During link failures, WRP has lo w er time complexit y than DSD V since it

nly

informs neigb

ring

no des ab

ut

link status c hanges. During link additions, hel lo messages are used as a presence indicator suc h that the routing table en try can b e up dated. Again, this

nly

aects neigh b

ring

no des. In CGSR, b ecause routing p erformance is dep enden t

n

the status

f

sp ecic no des (cluster head, gatew a y

r

normal no des), time complexit y

f

a link failure asso ciated with a cluster head is higher than DSD V, giv en the additional time needed to p erform cluster head reselection. Similarly , this applies to the case

f

link additions asso ciated with the cluster head. There is no gatew a y selection in CGSR since eac h no de declares it is a gatew a y no de to its neigh b

rs

if it is resp

nding

to m ultiple radio co des. If a gatew a y no de mo v es

ut
f

range, the routing proto col is resp

nsible

for routing the pac k et to another gatew a y . In terms

f

comm unication complexit y , since DSD V, CGSR and WRP use distance v ector shortest-path routing as the underlying routing proto col, they all ha v e the same degree

f

complexit y during link failures and additions. 1 While WRP itself uses at addressing, it can b e used hierarc hicall y [15]. 2 The proto col itself curren tly do es not supp

rt

m ulticast; ho w ev er, there is a separate proto col describ ed in [3], whic h runs

n

top

f

CGSR and pro vides m ulticast capabilt y . 11

SLIDE 12 P erformance P arameters A OD V DSR TORA ABR SSR Time Complexit y O(2d) O(2d) O(2d) O(d+z) O(d+z) (initializatio n) Time Complexit y O(2d) O(2d)

r

O(2d) O(l+z) O(l+z) (p

stfailure)

(cac he hit) Comm unication Complexit y O(2N) O(2N) O(2N) O(N+y) O(N+y) (initializatio n) Comm unication Complexit y O(2N) O(2N) O(2x) O(x+y) O(x+y) (p

stfailure)

Routing Philosoph y Flat Flat Flat Flat Flat Lo

p

F ree Y es Y es Y es Y es Y es Multicast Capabilit y Y es No No 3 No No Beaconing Requiremen ts No No No Y es Y es Multiple Route P

ssibilities

No Y es Y es No No Routes Main tained in route route route route route table cac he table table table Utilizes Route Cac he/T able Y es No No No No Expiration Timers Route Reconguratio n Erase Route; Erase Route; Link Rev ersal; Lo calized Erase Route; Metho dology Notify Source Notify Source Route Repair Broadcast Query Notify Source Routing Metric F reshest & Shortest Shortest P ath Asso ciativit y & Asso ciativit y & Shortest P ath P ath Shortest P ath & Stabilit y

thers

4 T able 2: Comparisons

f

the Characteristics

f

Source-Initiated On-Demand Ad-Ho c Routing Pro- to cols. Abbreviations: l = D iameter

f

the af f ected netw

r

k seg ment y = T

tal

number

f

nodes f

r

ming the dir ected path w her e the RE P LY pack et tr ansits z = D iameter

f

the dir ected path w her e the RE P LY pack et tr ansits 3.2 Source-Initiated On-Demand Routing Proto cols T able 2 presen ts a comparison

f

A OD V, DSR, TORA, ABR and SSR. The A OD V proto col emplo ys a route disco v ery pro cedure similar to DSR; ho w ev er, there are a couple imp

rtan

t dis- tinctions. The most notable

f

these is that the

v

erhead

f

DSR is p

ten

tially larger than that

f

A OD V since eac h DSR pac k et m ust carry full routing information, whereas in A OD V pac k ets need

nly

con tain the destination address. Similarly , the route replies in DSR are larger b ecause they con tain the address

f

ev ery no de along the route, whereas in A OD V route replies need

nly

carry the destination IP address and sequence n um b er. Also, the memory

v

erhead ma y b e sligh tly greater in DSR b ecause

f

the need to remem b er full routes, as

pp
sed

to

nly

next hop information in A OD V. A further adv an tage

f

A OD V is its supp

rt

for m ulticast [18 ]. None

f

the

ther

algorithms considered in this pap er curren tly incorp

rate

m ulticast comm unication. On the do wnside, A OD V requires symmetric links b et w een no des, and hence cannot utilize routes with assymetric links. In this asp ect, DSR is sup erior as it do es not require the use

f

suc h links, and can utilize assymetric links when symmetric links are not a v ailable. The DSR algorithm is in tended for net w

rks

in whic h the mobiles mo v e at a mo derate sp eed with resp ect to pac k et transmission latency [10 ]. Assumptions that the algorithm mak es for

p

eration are that the net w

rk

diameter is relativ ely small and that the mobile no des can enable a promiscuous receiv e mo de, whereb y ev ery receiv ed pac k et is deliv ered to the net w

rk

driv er soft w are without ltering b y destination address. An adv an tage

f

DSR

v

er some

f

the

ther
n-demand

proto cols 3 Lik e CGSR, TORA also do es not supp

rt

m ulticast; ho w ev er, there is a separate proto col, LAM [9], whic h runs

n

top

f

TORA and pro vides m ulticast capabilt y . 4 ABR also uses the Route Rela ying Load and Cum ulativ e F

rw

arding Dela y as routing metrics. 12

SLIDE 13 is that DSR do es not mak e use

f

p erio dic routing adv ertisemen ts, thereb y sa ving bandwidth and reducing p

w

er consumption. Hence the proto col do es not incur an y

v

erhead when there are no c hanges in net w

rk

top

logy

. Additionally , DSR allo ws no des to k eep m ultiple routes to a destination in their cac he. Hence, when a link

n

a route is brok en, the source no de can c hec k its cac he for another v alid route. If suc h a route is found, route reconstruction do es not need to b e rein v

k

ed. In this case, route reco v ery is faster than in man y

f

the

ther
n-demand

proto cols. Ho w ev er, if there are no additional routes to the destination in the source no de's cac he, route disco v ery m ust b e reinitiated, as in A OD V, if the route is still required. On the

ther

hand, b ecause

f

the small diameter assumption and b ecause

f

the source routing requiremen t, DSR is not scalable to large net w

rks.

F urthermore, as previously stated, the need to place the en tire route in b

th

r

ute

r eplies and data pac k ets causes greater con trol

v

erhead than in A OD V. TORA is a \link rev ersal" algorithm that is b est-suited for net w

rks

with large, dense p

pu-

lations

f

no des [16 ]. P art

f

the no v elt y

f

TORA stems from its creation

f

D A Gs to aid route establishmen t. One

f

the adv an tages

f

TORA is its supp

rt

for m ultiple routes. TORA and DSR are the

nly
n-demand

proto cols considered here whic h retain m ultiples route p

ssibilities

for a single source/destination pair. Route reconstruction is not necessary un til all kno wn routes to a destination are deemed in v alid, and hence bandwidth can p

ten

tially b e conserv ed b ecause

f

the necessit y for few er route rebuildings. Another adv an tage

f

TORA is its supp

rt

for m ulticast. Al- though, unlik e A OD V, TORA do es not incorp

rate

m ulticast in to its basic

p

eration, it functions as the underlying proto col for the Ligh t w eigh t Adaptiv e Multicast Algorithm (LAM), and together the t w

proto

cols pro vide m ulticast capabilit y [9 ]. TORA's reliance

n

sync hronized clo c ks, while a no v el idea, inheren tly limits its applicabilit y . If a no de do es not ha v e a GPS p

sitioning

system

r

some

ther

external time source, it cannot use the algorithm. Additionally , if the external time source fails, the algorithm will cease to

p

erate. F urther, route rebuilding in TORA ma y not

ccur

as quic kly as in the

ther

algorithms due to the p

ten

tial for

scillations

during this p erio d. This can lead to p

ten

tially length y dela ys while w aiting for the new routes to b e determined. ABR is a compromise b et w een broadcast and p

in

t-to-p

in

t routing and uses the connection-

rien

ted pac k et forw arding approac h. Route selection is primarily based

n

the aggregated asso ciativit y tic ks

f

no des along the path. Hence, although the resulting path do es not necessarily result in the smallest p

ssible

n um b er

f

hops, the path tends to b e longer-liv ed than

ther

routes. A long- liv ed route requires few er route reconstructions and therefore yields higher throughput. Another b enet

f

ABR is that, lik e the

ther

proto cols, it is guaran teed to b e free from pac k et duplicates. The reason is that

nly

the b est route is mark ed v alid while all

ther

p

ssible

routes remain passiv e. ABR, ho w ev er, relies

n

the fact that eac h no de is b eaconing p erio dically . The b eaconing in terv al m ust b e short enough so as to accurately reect the spatial, temp

ral,

and connectivit y state

f

the mobile hosts. This b eaconing requiremen t ma y result in additional p

w

er consumption. Ho w ev er, exp erimen tal results

btained

in [24 ] rev eal that the inclusion

f

p erio dic b eaconing has a min ute inuence

n

the

v

erall battery p

w

er consumption. Unlik e DSR, ABR do es not utilize route cac hes. The SSR algorithm is a logical descendan t

f

ABR. It utilizes a new tec hnique

f

selecting routes based

n

the signal strength and lo cation stabilit y

f

no des along the path. As in ABR, while the paths selected b y this algorithm are not necessarily shortest in hop coun t, they do tend to b e more stable and longer-liv ed, resulting in few er route reconstructions. One

f

the ma jor dra wbac ks

f

the SSR pro cotol is that, unlik e in A OD V and DSR, in termediate no des cannot reply to route requests sen t to w ards a destination; this results in p

ten

tially long dela ys b efore a route can b e disco v ered. Additionally , when a link failure

ccurs

along a path, the route disco v ery algorithm m ust b e re-in v

k

ed from the source to nd a new path to the destination. No attempt is made to use partial route reco v ery (unlik e ABR)

i.e.

to allo w in termediate no des to attempt to rebuild the route themselv es. A OD V and DSR also do not sp ecify in termediate no de rebuilding. While this ma y 13

SLIDE 14 P arameters On-Demand T able-Driv en Av ailabilit y

f

Av ailable Alw a ys a v ailable Routing Information when needed regardless

f

need Routing Flat Mostly at Philosoph y except for CSGR P erio dic route Not Y es up dates required Coping with Using lo calized Inform

ther

no des mobilit y route disco v ery to ac hiev e consisten t as in ABRand SSR routing table Signaling trac Gro ws with increasing Greater than generated mobilit y

f

activ e that

f
n-demand

routes (as in ABR) routing Qualit y

f

Service F ew can supp

rt

QoS Mainly Shortest P ath Supp

rt

as QoS metric T able 3: Ov erall Comparisons

f

On-Demand v ersus T able-Driv en Based Routing Proto cols. lead to longer route reconstruction times since link failures cannot b e resolv ed lo cally without the in terv en tion

f

the source no de, the attempt and failure

f

an in termediate no de to rebuild a route will cause a longer dela y then if the source no de had attempted the rebuilding as so

n

as the brok en link w as noticed. Th us it remains to b e seen whether in termediate no de route rebuilding is more

ptimal

than source no de route rebuilding. 3.3 T able-Driv en vs On-Demand Routing As discussed earlier, the table-driv en ad-ho c routing approac h is similar to the connectionless approac h

f

forw arding pac k ets, with no regard to when and ho w frequen t suc h routes are desired. It relies

n

an underlying routing table up date mec hanism that in v

lv

es the constan t propagation

f

routing information. This is, ho w ev er, not the case for

n-demand

routing proto cols. When a no de using an

n-demand

proto col desires a route to a new destination, it will ha v e to w ait un til suc h a route can b e disco v ered. On the

ther

hand, b ecause routing information is constan tly propagated and main tained in table-driv en routing proto cols, a route to ev ery

ther

no de in the ad-ho c net w

rk

is alw a ys a v ailable, regardless

f

whether

r

not it is needed. This feature, although useful for datagram trac, incurs substan tial signaling trac and p

w

er consumption. Since b

th

bandwidth and battery p

w

er are scarce resources in mobile computers, this b ecomes a serious limitation. T able 3 lists some

f

the basic dierences b et w een the t w

classes
f

algorithms. Another consideration is whether a at

r

hierarc hical addressing sc heme should b e used. All

f

the proto cols considered here, except for CGSR, use a at addressing sc heme. In [1], a discussion

f

the t w

addressing

sc hemes is presen ted. While at addressing ma y b e less complicated and easier to use, there are doubts as to its scalabilit y . 4 Applications and Challenges Akin to pac k et radio net w

rks,

ad-ho c wireless net w

rks

ha v e an imp

rtan

t role to pla y in military applications. Soldiers equipp ed with m ulti-mo de mobile comm unicators can no w comm unicate in an ad-ho c manner, without the need for xed wireless base stations. In addition, small v ehicular devices equipp ed with audio sensors and cameras can b e deplo y ed at targetted regions to collect imp

rtan

t lo cation and en vironmen tal information whic h will b e comm unciated bac k to a pro cessing no de via ad-ho c mobile comm unications. Ship-to-ship ad-ho c mobile comm unication is also desirable since it 14

SLIDE 15 pro vides alternate comm unication paths without reliance

n

ground-

r

space-based comm unication infrastructures. Commerical scenarios for ad-ho c wireless net w

rks

include: (a) conferences/meetings/lectures [8], (b) emergency services, and (c) la w enforcemen t. P eople to da y attend meetings and conferences with their laptops, palm tops and noteb

ks.

It is therefore attractiv e to ha v e instan t net w

rk

formation, in addition to le and information sharing without the presence

f

xed base stations and systems administrators. A presen ter can m ulticast slides and audio to in tended receip en ts. A ttendees can ask questions and in teract

n

a commonly-shared white b

ard.

Ad-ho c mobile comm unication is particularly useful in rela ying information (status, situation a w areness, etc.) via data, video and/or v

ice

from

ne

rescue team mem b er to another

v

er a small handheld

r

w earable wireless device. Again, this applies to la w enforcemen t p ersonnel as w ell. Curren t c hallenges for ad-ho c wireless net w

rks

include: (a) m ulticast, (b) QoS supp

rt,

(c) p

w

er-a w are routing [20 ], and (d) lo cation-aided routing [12 ]. As men tioned ab

v

e, m ulticast is desirable to supp

rt

m ulti-part y wireless comm unications. Since the m ulticast tree is no longer static (i.e., its top

logy

is sub ject to c hange

v

er time), the m ulticast routing proto col m ust b e able to cop e with mobilit y , including m ulticast mem b ership dynamics (suc h as l eav e and j

in).

In terms

f

QoS, it is inadequate to consider QoS merely at the net w

rk

lev el without considering the underlying media access con trol la y er [13]. Again, giv en the problems asso ciated with the dynamics

f

no des, hidden terminals, and uctuating link c haracteristics, supp

rting

end-to-end QoS is a non- trivial issue that requires in-depth in v estigation. Curren tly , there is a trend to w ards an adaptiv e QoS approac h instead

f

the \plain" resource reserv ation metho d with hard QoS guaran tees. Another imp

rtan

t factor is the limited p

w

er supply in handheld devices whic h can seriously prohibit pac k et forw arding in an ad-ho c mobile en vironmen t. Hence, routing trac based

n

no des' p

w

er metric is

ne

w a y to distinguish routes that are more long-liv ed than

thers.

Finally , instead

f

using b eaconing

r

broadcast searc h, lo cation-aided routing uses p

sitioning

information to dene asso ciated regions so that the routing is spatially-orien ted and limited. This is analogous to asso ciativit y-orien ted and restricted broadcast in ABR. Curren t ad-ho c routing approac hes ha v e in tro duced sev eral new paradigms, suc h as exploit- ing user's demand, the use

f

lo cation, p

w

er, and asso ciation parameters. Adaptivit y and self- conguration are k ey features

f

these approac hes. Ho w ev er, exibilit y is also imp

rtan

t. A f l exibl e ad-ho c routing proto col could resp

nsiv

ely in v

k

e table-driv en approac hes and/or

n-demand

approac hes based

n

situations and comm unication requiremen ts. The \toggle" b et w een these t w

approac

hes ma y not b e trivial since concerned no des m ust b e \in-sync" with the toggling. Co- existence

f

b

th

approac hes ma y also exist in spatially clustered ad-ho c groups, with in tra-cluster emplo ying the table-driv en approac h and in ter-cluster emplo ying the demand-driv en approac h

r

vice v ersa. F urther w

rk

is necessary to in v estigate the feasibilit y and p erformance

f

h ybrid ad-ho c routing approac hes. Lastly , in addition to the ab

v

e, further researc h in the areas

f

media access con trol, securit y , service disco v ery , and in ternet proto col

p

erabilit y is required b efore the p

ten

tial

f

ad-ho c mobile net w

rking

can b e realized. 5 Conclusion In this pap er w e ha v e pro vided descriptions

f

sev eral routing sc hemes prop

sed

for ad-ho c mobile net w

rks.

W e ha v e also pro vided a classication

f

these sc hemes according to the routing strategy , i.e., table-driv en and

n-demand.

W e ha v e presen ted a comparison

f

these t w

categories
f

routing proto cols, highligh ting their features, dierences and c haracteristics. Finally , w e ha v e iden tied p

ssible

applications and c hallenges facing ad-ho c mobile wireless net w

rks.

While it is not clear 15

SLIDE 16 that an y particular algorithm

r

class

f

algorithm is the b est for all scenarios, eac h proto col has denite adv an tages and disadv an tages and has certain situations for whic h it is w ell-suited. The eld

f

ad-ho c mobile net w

rks

is rapidly gro wing and c hanging, and while there are still man y c hallenges that need to b e met, it is lik ely that suc h net w

rks

will see wide-spread use within the next few y ears. 6 Ac kno wledgmen t W e w

uld

lik e to thank C.-C. Chiang, J. J. Garcia-Luna-Acev es, Da vid Maltz, Vincen t P ark, and Charles P erkins for their assistance in assuring accurate descriptions

f

the

p

eration

f

their protocols. References [1] D. Bak er, M. S. Corson, P . Sass, and S. Ramanatham, \Flat vs. Hierarc hical Net- w

rk

Con trol Arc hitecture," AR O/D ARP A W

rkshop
n

Mobile Ad-Ho c Net w

rking,

h ttp://www.isr.umd.edu/Courses/W

rkshops/MANET/pro

gram. h tml, Marc h 1997. [2] J. Bro c h, D. B. Johnson, D. A. Maltz, \The Dynamic Source Routing Proto col for Mobile Ad Ho c Net w

rks,"

IETF Internet Dr aft draft-ietf-manet-dsr-01.txt, Decem b er 1998 (W

rk

in Progress). [3] C.-C. Chiang, M. Gerla, and L. Zhang, \Adaptiv e Shared T ree Multicast in Mobile Wireless Net w

rks,"

Pr

c

e e dings

f

GLOBECOM '98, pp. 1817{1822, No v em b er 1998. [4] C.-C. Chiang, H.K. W u, W. Liu, and M. Gerla, \Routing in Clustered Multihop, Mobile Wireless Net w

rks

with F ading Channel," Pr

c

e e dings

f

IEEE SICON'97, pp. 197{211, April 1997. [5] M. S. Corson and A. Ephremides, \A Distributed Routing Algorithm for Mobile Wireless Net- w

rks,"

A CM/Baltzer Wir eless Networks Jouornal, V

l.

1, No. 1, pp. 61{81, F ebruary 1995. [6] R. Dub e, C. D. Rais, K.-Y. W ang, and S.K. T ripathi, \Signal Stabilit y based Adaptiv e Routing (SSA) for Ad-Ho c Mobile Net w

rks,"

IEEE Personal Communic ations, pp. 36{45, F ebruary 1997. [7] L. R. F

rd

Jr. and D. R. F ulk erson, Flows in Networks. Princeton Univ ersit y Press, Princeton N.J., 1962. [8] M. Gerla, C.-C. Chiang, and L. Zhang, \T ree Multicast Strategies in Mobile, Multihop Wireless Net w

rks,"

T

app

ear in A CM/Baltzer Mobile Networks and Applic ations Journal, 1998. [9] L. Ji and M. S. Corson, \A Ligh t w eigh t Adaptiv e Multicast Algorithm," Pr

c

e e dings

f

GLOBE- COM '98, pp. 1036{1042, No v em b er 1998. [10] D. B. Johnson and D. A. Maltz, \Dynamic Source Routing in Ad-Ho c Wireless Net w

rks,"

Mobile Computing, ed. T. Imielinski and H. Korth, Klu w er Academic Publishers, pp. 153{181, 1996. [11] J. Jubin and J. T

rno

w, \The D ARP A P ac k et Radio Net w

rk

Proto cols," Pr

c

e e dings

f

the IEEE, V

l.

75, No. 1, pp. 21{32, 1987. 16

SLIDE 17 [12] Y. B. Ko and N. H. V aidy a, \Lo cation-Aided Routing (LAR) in Mobile Ad Ho c Net w

rks,"

Pr

c

e e dings

f

A CM/IEEE MOBICOM '98, Octob er 1998. [13] C. R. Lin and M. Gerla, \MA CA/PR: An Async hronous Multimedia Multihop Wireless Net- w

rk,"

Pr

c

e e dings

f

IEEE INF OCOM '97, Marc h 1997. [14] S. Murth y and J. J. Garcia-Luna-Acev es, \An Ecien t Routing Proto col for Wireless Net- w

rks,"

A CM Mobile Networks and Applic ations Journal, Sp ecial Issue

n

Routing in Mobile Comm unication Net w

rks,

pp. 183{197, Octob er 1996. [15] S. Murth y and J. J. Garcia-Luna-Acev es, \Lo

p-F

ree In ternet Routing Using Hierarc hical Rout- ing T rees," Pr

c

e e dings

f

INF OCOM '97, April 7-11, 1997. [16] V. D. P ark and M. S. Corson, \A Highly Adaptiv e Distributed Routing Algorithm for Mobile Wireless Net w

rks,"

Pr

c

e e dings

f

INF OCOM '97, April 1997. [17] C. E. P erkins and P . Bhagw at, \Highly Dynamic Destination-Sequenced Distance-V ector Rout- ing (DSD V) for Mobile Computers," Computer Communic ations R eview, pp. 234{244, Octob er 1994. [18] C. E. P erkins and E. M. Ro y er, \Ad Ho c On Demand Distance V ector (A OD V) Routing," IETF Internet Dr aft, draft-ietf-manet-ao dv-02.txt, No v em b er 1998 (W

rk

in Progress). [19] C. E. P erkins and E. M. Ro y er, \Ad-ho c On-Demand Distance V ector Routing," Pr

c

e e dings

f

2nd IEEE Workshop

n

Mobile Computing Systems and Applic ations, F ebruary 1999. [20] S. Singh, M. W

,

and C. S. Ragha v endra, \P

w

er-Aw are Routing in Mobile Ad Ho c Net w

rks,"

Pr

c

e e dings

f

A CM/IEEE MOBICOM '98, Octob er 1998. [21] A. S. T anen baum, Computer Networks, Thir d Edition. Pren tice Hall, Englew

d

Clis, Chapter 5, pp. 357{358, 1996. [22] C-K. T

h,

\A No v el Distributed Routing Proto col T

Supp
rt

Ad-Ho c Mobile Computing," Pr

c

e e dings

f

the 1996 IEEE Fifte enth A nnual International Pho enix Confer enc e

n

Computers and Communic ation, pp. 480{486, Marc h 1996. [23] C-K. T

h,

\Asso ciativit y-Based Routing for Ad-Ho c Mobile Net w

rks,"

Wir eless Personal Com- munic ations, V

l.

4, No. 2, pp. 1{36, Marc h 1997. [24] C-K. T

h

and George Lin, \Implemen ting Asso ciativit y-Based Routing for Ad Ho c Mobile Wireless Net w

rks,"

Unpublished Article, Marc h 1998. 17

SLIDE 18 Authors' Biographies ELIZABETH M. R O YER (ero y er@alpha.ece.ucsb.edu) receiv ed her M.S. degree in Electrical and Com- puter Engineering from the Univ ersit y

f

California, San ta Barbara in Decem b er 1997 and her B.S. degrees in Computer Science and Applied Mathematics from Florida State Univ ersit y in April 1996. She is curren tly a PhD studen t in the Electrical and Computer Engineering Departmen t at the Univ ersit y

f

California, San ta Barbara. A t UCSB she is a mem b er

f

the Computer Net w

rks

and Distributed Systems Lab

ratory

. Her researc h in terests include routing and m ulticast for ad-ho c mobile net w

rks.

Elizab eth is the recipien t

f

a National Science F

undation

Graduate F ello wship and a Univ ersit y

f

California Do ctoral Sc holar F ello wship. CHAI-KEONG TOH (C-K.T

h@acm.org)

w as b

rn

in Singap

re

in 1965. He receiv ed his Diploma in Electronics and Comm uni cation Engineering with a Certicate

f

Merit a w ard from the Singap

re

P

lytec

hnic in 1986, his BEng degree in Electronics Engineering with rst class honors from the Univ ersit y

f

Manc hester Institute

f

Science and T ec hnology (UMIST) in 1991 and his Ph.D. degree in Computer Science from the Computer Lab

ratory

, Univ ersit y

f

Cam bridge, Cam bridge, England in 1996. Dr. T

h

founded and c haired the Mobile Sp ecial In terest Group (Mobile SIG) from 1994-1996. Before joining Cam bridge Univ ersit y , C-K w as a net w

rk

sp ecialist, R&D engineer and tec hnical sta mem b er. A t Cam bridge, C-K w as an Honorary Cam bridge Commo n w ealth T rust Sc holar and a King's College Cam bridge Researc h Sc holar. He authored the b

k
n

`Wireless A TM and Ad Ho c Net w

rks:

Proto cols and Arc hitectures', whic h w as published b y Klu w er Academic Press in 1996. Dr. T

h

serv es

n

the editorial b

ard

for A CM Mobile Computing and Comm uni cations Review (MC2R), IEEE Net w

rk

and P ersonal T ec hnologies Journal (Springer V erlag). He is a mem b er

f

IEE, IEEE, USENIX, A CM, Sigma Xi Honor So ciet y , New Y

rk

Academ y

f

Science, American Asso ciation for the Adv ancemen t

f

Science (AAAS) and is a F ello w

f

the Cam bridge Philosophical So ciet y and Cam bridge Common w eal th So ciet y . He is curren tly an assistan t professor with the Sc ho

l
f

Electrical and Computer Engineering at the Georgia Institute

f

T ec hnology , A tlan ta, Georgia, directing the Mobile Multimedia and High Sp eed Net w

rking

Lab

ratory

. 18