Intro duction
Laptop computers
to show
capacity and avail
continue convenience mobility memory
improvements
Adho c OnDemand Distance Vector Routing
Charles E Perkins
Sun Microsystems Lab oratories Advanced Development Group Menlo Park CA
cp erkinsengsuncom
Abstract
An adho c network is the cooperative
a col lection of mobile nodes without the
vention of any centralized access point
frastructure In this paper we present Adhoc On De
mand Distance Vector Routing
rithm for the operation of such adhoc networks Each Mobile Host operates as a specialized router and routes are obtained as needed ie ondemand with little or no reliance on periodic advertisements Our new routing algorithm is quite suitable for a dynamic self starting network as required by users wishing to utilize adhoc networks AODV provides loopfree routes even while repairing broken links Because the protocol does not require global periodic routing advertisements the demand on the overal l bandwidth available to the mo bile nodes is substantial ly less than in those protocols that do necessitate such advertisements Nevertheless we can stil l maintain most of the advantages of basic distancevector routing mechanisms We show that our algorithm scales to large populations of mobile nodes wishing to form adhoc networks We also include an evaluation methodology and simulation results to verify the operation of our algorithm
Keywords Adho c Networking Distance Vector Routing Dynamic Routing Mobile Networking Wire less Networks
in
ability of disk storage These smaller computers can b e equipp ed with gigabytes of disk storage high res olution color displays p ointing devices and wireless communications adapters Moreover b ecause many of these small in size only computers op erate with bat
engagement required inter or existing in
AODV a novel algo
Elizab eth M Royer
Dept of Electrical and Computer Engineering University of California Santa Barbara Santa Barbara CA eroyeralphaeceucsbedu
tery p ower users are free to move ab out at their con venience without b eing constrained by wires
The idea of forming an onthey adho c network of mobile no des dates back to DARPA packet radio net work days More recently the interest in this sub ject has grown due to availability of licensefree wireless communication devices that users of laptop computers can use to communicate with each other Several recent pap ers on this topic have fo cused on the algorithmic complexity of cho osing the optimal set of adho c routers while others have prop osed new routing solutions leverag ing features from the existing Internet routing algo rithms Interest within the Internet Engineering Task Force IETF is also growing as is evidenced by the for mation of a new working group manet whose charter is to develop a solution framework for rout ing in adho c networks The manet working group has goals that are quite distinct from the goals of the IETF mobileip working group and make little or no use of Mobile IP or any of its forerunners eg
The DestinationSequenced Distance Vector DSDV algorithm has b een prop osed as a variant of the distance vector routing metho d by which mobile no des co op erate to form an adho c network DSDV is eective for creating adho c networks for small p opu lations of mobile nodes but it is a fairly brute force approach because it depends for its correct operation on the periodic advertisement and global dissemina tion of connectivity information Frequent systemwide broadcasts limit the size of adho c networks that can eectively use DSDV b ecause the control message over head grows as O n DSDV also requires each mobile no de to maintain a complete list of routes one for each destination within the adho c network This almost al ways exceeds the needs of any particular mobile no de Keeping a complete routing table do es reduce route ac quisition latency b efore transmission of the rst packet to a destination It is however p ossible to design a sys
of
tem whereby routes are created ondemand eg Such systems must take steps to limit the time used for route acquisition otherwise users of the adho c no des might exp erience unacceptably long waits b efore trans mitting urgent information The advantage here is that a smo othly functioning adho c system with ondemand routes could largely eliminate the need for p erio dic broadcast of route advertisements With the goals of minimizing broadcasts and transmission latency when new routes are needed we designed a proto col to im prove up on the p erformance characteristics of DSDV in the creation and maintenance of adho c networks
Although AODV do es not dep end sp ecically on particular asp ects of the physical medium across which
nicate unless the former node is oering its services as an intermediate forwarding station to maintain con nectivity b etween two other no des
When the lo cal connectivity of the mobile no de is of interest each mobile no de can b ecome aware of the other nodes in its neighborhood by the use of several techniques including local not systemwide broad casts known as hello messages The routing tables of the no des within the neighb orho o d are organized to op timize resp onse time to lo cal movements and provide quick resp onse time for requests for establishment of new routes The algorithms primary objectives are
To broadcast discovery packets only when neces sary
To distinguish b etween lo cal connectivity manage
packets
largely
such as
wireless
a mobile no de can have neighb ors which hear its broad casts and yet do not detect each other the hidden ter minal problem We do not make any attempt to
are disseminated its development has b een motivated by limited range broadcast media those utilized by infrared or radio frequency communications adapters Using such media
ment neighborhood detection and
ogy maintenance
To disseminate information ab out
general top ol
use
our
by
ters No des that need to op erate over multiple chan nels are presumed to be able to do so The algorithm works on wired media as well as wireless media as long as links along which packets may b e transmitted are
that are likely to need the information
AODV uses a broadcast route discovery mecha nism as is also used with mo dications in the Dy namic Source Routing DSR algorithm Instead of source routing however AODV relies on dynam ically establishing route table entries at intermediate no des This dierence pays o in networks with many no des where a larger overhead is incurred by carry ing source routes in each data packet To maintain the most recent routing information b etween no des we b orrow the concept of destination sequence num b ers from DSDV Unlike in DSDV however each adho c no de maintains a monotonical ly increasing se quence numb er counter which is used to sup ersede stale cached routes The combination of these techniques yields an algorithm that uses bandwidth eciently by minimizing the network load for control and data traf c is resp onsive to changes in top ology and ensures lo opfree routing
2.1. Path Discovery
sp ecic characteristics of the physical medium in
algorithm nor to handle sp ecic problems p osed channelization needs of radio frequency transmit
available The only requirement cast medium is that neighb oring others broadcasts
placed on no des can
the broad detect each
AODV uses
no des It do es
no des when one of the no des cannot hear the other one however we may include the use of such links in future enhancements Steps to prevent use of such asymmet ric links b etween no des are describ ed briey in Sec tion
symmetric links not attempt to
b etween
follow paths b etween
The remainder of this
In Section the proto col details
Section presents the simulations input parameters and results obtained Section describ es our plans for future work and nally Section concludes the pap er
The Adho c OnDemand Vector Algorithm
Our basic prop osal can b e
Distance
pap er is organized as follows for AODV are given
route acquisition system
tive paths neither maintain any routing information nor participate in any p erio dic routing table exchanges Further a no de do es not have to discover and maintain a route to another no de until the two need to commu
the
following elds
neighb oring
ondemand
RREQ packet to its
neighb ors
The RREQ
contains
changes in lo cal connectivity to those neighb oring mobile no des
The Path
source no de
for which it has no routing information in its table
Discovery pro cess is initiated whenever a needs to communicate with another no de
Every no de
sequence number and a broadcast id The source no de initiates path discovery by broadcasting a route request
maintains two separate counters a node
called a pure
no des that do not lie on ac
sour ce seq uence br oadcast id dest addr dest seq uence hop cnt
sour ce addr
The pair sour ce addr br oadcast id uniquely identies a RREQ br oadcast id is incremented when ever the source issues a new RREQ Each neighb or either satises the RREQ by sending a route reply RREP back to the source see Section or re broadcasts the RREQ to its own neighb ors after in creasing the hop cnt Notice that a no de may receive multiple copies of the same route broadcast packet from
As the RREQ travels from a source to various desti nations it automatically sets up the reverse path from
all no des back to the source
ure To set up a reverse
address of the neighb or from which it received the rst copy of the RREQ These reverse path route entries are maintained for at least enough time for the RREQ to traverse the network and pro duce a reply to the sender
2.1.2. Forward Path Setup
as illustrated in Fig path a no de records the
various neighb ors
a RREQ if it has
same br oadcast id and source address it drops the re dundant RREQ and do es not rebroadcast it If a no de cannot satisfy the RREQ it keeps track of the follow
When an intermediate no de receives already received a RREQ with the
ing information in order to implement the setup as well as the forward path setup
reverse path
company the transmission of the
eventual
Destination IP address
Source IP address
Broadcast id
Expiration time for reverse path route entry
that will RREP
ac
RREQ was received over a bidirectional intermediate no de has a route entry for the
link If desired
Eventually a RREQ will arrive at a
the destination itself that p ossesses a current route to the destination The receiving no de rst checks that
the
an
destination it determines whether the route is current by comparing the destination sequence numb er in its own route entry to the destination sequence numb er in the RREQ If the RREQs sequence numb er for the destination is greater than that recorded by the inter mediate no de the intermediate no de must not use its recorded route to resp ond to the RREQ Instead the
no de
p ossibly
Source
no des sequence
D
numb er
D
intermediate no de
mediate node can reply only when it has a route with a sequence numb er that is greater than or equal to that contained in the RREQ If it do es have a current route to the destination and if the RREQ has not b een pro cessed previously the no de then unicasts a route re
rebroadcasts the RREQ The inter
ply packet RREP back to its received the RREQ A RREP information
neighb or from which it contains the following
sour ce addr dest addr dest seq uence hop cnt l if etime
Figure 1. Reverse Path Formation
2.1.1. Reverse Path Setup
Figure 2. Forward Path Formation
There are two sequence
the broadcast id included in a RREQ the source se quence numb er and the last destination sequence num b er known to the source The source sequence numb er is used to maintain freshness information about the re verse route to the source and the destination sequence numb er sp ecies how fresh a route to the destination must b e b efore it can b e accepted by the source
numb ers in
addition
to
arrives at a no de destination a re to the source of the RREP travels back no de along the path sets up a the no de from which the RREP timeout information for route en tries to the source and destination and records the latest destination sequence numb er for the requested destination Figure represents the forward path setup as the RREP travels from the destination D to the source no de S No des that are not along the path determined by the RREP will timeout after AC TIVE ROUTE TIMEOUT msec and will delete
p ointers
receiving an RREP propagates the rst
a given source no de towards that source If it receives further RREPs it up dates its routing in formation and propagates the RREP only if the RREP
S
timeout
By the time a broadcast packet that can supply a route to the
S
verse path has b een established RREQ Section As the to the source each
forward p ointer to
came up dates its
the reverse A no de RREP for
contains either a greater destination sequence numb er than the previous RREP or the same destination se quence numb er with a smaller hop count It suppresses all other RREPs it receives This decreases the num b er of RREPs propagating towards the source while also ensuring the most uptodate and quickest routing information The source no de can b egin data trans mission as soon as the rst RREP is received and can later update its routing information if it learns of a b etter route
2.2. Route Table Management
Each time a route entry is used to transmit data from a source toward a destination the timeout
which the route is considered to b e invalid routing table entry the address of active through which packets for the given desti received is also maintained A neighb or is considered active for that destination if it origi nates or relays at least one packet for that destination within the most recent activ e timeout p erio d This in formation is maintained so that all active source no des can be notied when a link along a path to the des tination breaks A route entry is considered active if it is in use by any active neighb ors The path from a source to a destination which is followed by pack ets along active route entries is called an active path Note that as with DSDV all routes in the route table are tagged with destination sequence numb ers which guarantee that no routing lo ops can form even under extreme conditions of outoforder packet delivery and
metric
destination route
for the entry is reset to the current time
plus ac
the mo numb er numb er
for the current route The route with the greater se quence numb er is chosen If the sequence numb ers are the same then the new route is selected only if it has
tive If
bile
route timeout
a new route is no de compares
In addition to the source and
numb ers other useful information is also stored in the route table entries and is called the softstate asso ciated with the entry Asso ciated with reverse path routing entries is a timer called the route request ex piration timer The purp ose of this timer is to purge reverse path routing entries from those no des that do not lie on the path from the source to the destination The expiration time dep ends up on the size of the ad ho c network Another imp ortant parameter asso ciated with routing entries is the route caching timeout or the
time after In each
neighb ors nation are
destination
sequence
oered to a mobile no de the destination sequence of the new route to the destination sequence
a smaller metric fewer numb er of nation
2.3. Path Maintenance
hops to the
desti
Movement of no des not lying
does not aect the routing to that paths destination If the source no de moves during an active session it can reinitiate the route discovery pro cedure to estab lish a new route to the destination When either the destination or some intermediate no de moves a sp ecial RREP is sent to the aected source nodes Periodic hel lo messages can b e used to ensure symmetric links as well as to detect link failures as describ ed in Sec tion Alternatively and with far less latency such failures could b e detected by using linklayer acknowl edgments LLACKS A link failure is also indicated if attempts to forward a packet to the next hop fail
Once the next hop b ecomes unreachable the no de upstream of the break propagates an unsolicited RREP with a fresh sequence numb er ie a sequence numb er that is one greater than the previously known sequence numb er and hop count of to all active upstream neighb ors Those no des subsequently relay that mes sage to their active neighb ors and so on This pro cess continues until all active source no des are notied it terminates b ecause AODV maintains only lo opfree routes and there are only a nite numb er of no des in the adho c network
Up on receiving notication of a broken link source no des can restart the discovery pro cess if they still re quire a route to the destination To determine whether
route
level
tions
the source no de or any other no de along the previ ous route decides it would like to rebuild the route to the destination it sends out an RREQ with a desti nation sequence numb er of one greater than the previ ously known sequence numb er to ensure that it builds a new viable route and that no nodes reply if they still regard the previous route as valid
has b een
proto col
remain op en using the
used recently control blo cks
as well as insp ect upp er to see whether connec indicated destination If
along an
active path
high no de mobility see App endix A
A mobile no de maintains a route table entry for each
destination of interest Each route table entry contains the following information
Destination
Next Hop
Numb er of hops
Sequence numb er for the
Active neighb ors for this
Expiration time for the route table entry
a route is still
needed a no de may check whether the
S DATA
VOICE
Simulated proto col
Packet size bytes
Packet count
Interarrival time of data packets Session interval sec
UDP
Exp onentialmean msec Geometricmean
UDP
Exp onentialmean msec Geometricmean
2.4. Local Connectivity Management
No des learn of their neighb ors in one of two ways
develop ed at UCLA as the successor to Maisie The PARSEC language is suited to the simulation of dy namic top ologies and routing algorithms
The main ob jective of our simulations is to show that ondemand route establishment with AODV is both quick and accurate Additional objectives include showing that AODV scales well to large networks and determining the optimal value for each of the necessary parameters
3.1. Simulation Environment
Our simulations were run using networks of and no des The movement algorithm for all network sizes is the same No des are initially placed randomly within a xedsize L L area During the simulation no des are free to move anywhere within this area Each no de cho oses a sp eed from a uniform distri bution b etween and meters p er second It then travels towards a random sp ot within the L L area The no de moves until it reaches that sp ot then cho oses a rest p erio d from a uniform distribution b etween
from a neigh information to In the event all of its ac interval it broadcasts to its neighb ors a hel lo message a sp e cial unsolicited RREP containing its identity and sequence numb er The no des sequence numb er is not changed for hel lo message transmissions This
Whenever a no de receives a broadcast b or it up dates its lo cal connectivity
ensure
that a
tive downstream neighb ors within hello
that it includes this neighb or no de has not sent any packets to
sages are not received
tive path the active
are sent notication
Section We have determined the for allowed hello loss is two as is tion
seconds After the rest p erio d the no de travels another randomly selected sp ot This pro cess throughout the simulation causing continuous
Table 1. Session-Dependent Traffic Parameters.
hel lo
side
tains
that
ity information
or a hel lo from
allowed hello loss consecutive hel lo messages from a no de previously in the neighb orho o d is an indica tion that the lo cal connectivity has changed Fail ing to receive hel lo messages from inactive neighbors do es not trigger any proto col action If hel lo mes
message is prevented from being rebroadcast out the neighb orho o d of the no de b ecause it con
a time to live TTL value of receive this packet up date their lo cal to the no de Receiving a
Neighb ors connectiv broadcast a new neighbor or failing to receive
from the next hop along an ac neighb ors using that next hop of link failure as describ ed in
optimal value shown in Sec
The local connectivity management with hel lo mes
also uses the same channel transmission carrier sens ing is p erformed by a no de to determine whether any of its neighb ors is transmitting If the no de detects an ongoing transmission by a neighb or it calculates an ex p onential backo based on the numb er of times it has attempted the retransmission and waits this amount of
and
towards
rep eats
changes in the top ology of the underlying network
Each of the simulations
mo del
time b efore listening to the tempts to transmit a packet dropping the packet
Before
b eginning a
sages can also b e used to
bidirectional connectivity are considered to b e neigh b ors For this purp ose each hel lo sent by a no de lists the no des from which it has heard Each no de checks to make sure that it uses only routes to neighb ors that have heard the nodes hel lo message To save local
ensure that only no des with
We have simulated AODV using an eventdriven packetlevel simulator called PARSEC which was
simultaneously
packets collide at no de B and b oth packets are dropp ed
Each no de creates a session to another no de selected at random The sessions created for each simulation
channel again A no de at max retrans times b efore
bandwidth such checking should b e p erformed explicitly congured into the no des
Simulations and Results
only if
No des in the
hidden terminal
B and no de C unable to hear no de As transmission
simulation
problem If no de A transmits to no de
frequently suer from the
transmits to no de B we assume the
are of
S DATA packets or voice
each of the session typ es are given in Table We chose to use the small data packet sessions for most of our simulations b ecause the larger size of the voice packets and the greater numb er of sessions generated tended to congest the network and hence decrease the go o dput ratio Nevertheless we include the results from these simulations to oer a contrast to the lighter demands
homogeneous
of the small data packets and to place a greater on the proto col
typ e
they are
data The parameters for
A session sends data segments until
sent the desired numb er of segments or it receives a
either
small data
either
it has
stress
we did not
The interconnection pattern of an adho c network is determined in part by the communication range Rmax For our simulations we held Rmax constant at m Two no des can communicate directly and are
Hello Interval
Route Discovery Timeout
Route Expiration
Reverse Route Life
Maximum of Retransmissions
msec
msec
msec
msec
thus considered each others neighb ors
than Rmax distance apart The ro om
and no des networks is mm
we found mm to b e to o small so we increased the dimensions to mm Similarly for no des we used a ro om size of mm
Table gives the values of the essential parameters for AODV The parameter values were chosen b ecause
3.2. Results and Discussion
and allowed hello loss we varied
ob jective was to show quickly and accurately at this time know an optimal value
that AODV can
Our rst nd routes
Since
for rreq retries rreq retries b e
if they are less size for the For no des
Table 2. Simulated Parameter Values
they minimize network congestion while
allowing the accurately as
algorithm to p ossible
op erate
as quickly
and as
a we intuitively guessed would b e reasonable Fig ure shows the go o dput ratios for and no des
tween value
and and set allowed hello loss to
timeout message from the network layer
triggered when a no de has sent a RREQ for a particular destination and has not received a valid route within
available during a session packets are network layer The data rate for b oth Mbitsec
Timeouts are
using
go o dput ratio is consistently ab ove For no des the go o dput ratio for rreq retries is ap proximately but then it decreases to for rreq retries and then increases with increasing values of rreq retries Bro ch et al simulated AODV over a network of no des and achieved go o d put ratios b etween and dep ending on the amount of time the no des were stationary during the simulation Note that our S DATA simulation uses the same size data packets as they did Hence our
the S DATA session typ e For no des the
route discovery timeout Any time a route
is not by the session typ es is
dropp ed
Each simulation is run for seconds and new ses sions are generated throughout the simulation Hence
we keep track of and account for
any uncompleted at the end of the
sessions and data simulation
packets in transit
100
99
98
97
96
95
94
93
92
91
90
50 Nodes 100 Nodes
0 0.5 1 1.5 2 2.5 3
RREQ_RETRIES
100
99
98
97
96
95
94
93
92
91
90
50 Nodes 100 Nodes
0 0.5 1 1.5 2 2.5 3 ALLOWED_HELLO_LOSS
Figure 3. Achieved Goodput for Varying
rreq retries
Figure 4. Achieved Goodput for Varying
allowed hello loss
Goodput(%)
Goodput(%)
300
250
200
150
100
50
0
50 Nodes 100 Nodes
0 0.5 1 1.5 2 2.5 3 RREQ_RETRIES
600
500
400
300
200
100
0
50 Nodes 100 Nodes
0 0.5 1 1.5 2 2.5 3 ALLOWED_HELLO_LOSS
Figure 5. Route Acquisition Latency for Varying rreq retries
achieved go o dput ratio for a no de network roughly corresponds with their results for the same size net work with our results b eing slightly b etter We disre gard the articially high go o dput ratio for no des and rreq retries b ecause more of the ses sions ab orted in this simulation than in the simulations with larger rreq retries values Given the remaining go o dput ratios for and no des we set the optimal rreq retries value to
Figure 6. Route Acquisition Latency for Varying allowed hello loss
rst RREQ was sent If a route to a destination was never found this time lapse was not taken into account in the computation Figure shows the computed route acquisition latencies for varying rreq retries values and Figure shows the corresp onding values for varying allowed hello loss values With the exception of rreq retries the minimum route acquisition latency was attained for the combination rreq retries allowed hello loss giving fur ther credence to our choice of parameter values
We with
hello
then simulated rreq retries loss parameter
and no des networks
the allowed of these simu lations are shown in Figure Here for no des allowed hello loss pro duced the b est results while for no des allowed hello loss was the b est Again b ecause is an unrealistic value and b ecause allowed hello loss pro duced the second
to b e
the essential results of our simulations
and varied The results
Table gives
for networks of and no des The re sults were obtained using the S DATA session typ e and setting rreq retries and allowed hello loss The bandwidth overhead ratio is a metric taken from although there it is called bandwidth utilization and is computed by dividing the total numb er of bits
b est results
the optimal
nding that
ter p erformance In their simulations
numb er of data bits trans
we chose
value This contradicts
allowed hello loss Bro ch
calculation
b ecause it
gives over b oth
transmitted by the total
mitted We include this
a go o d representation of the amount of control
head asso ciated with a proto col We also rep ort
the instantaneous go o dput ratio at the simulation end as well as the average go o dput ratio throughout the simulation b ecause these numb ers can vary due to a sudden increase in link breakages or session cre ations at the end of the simulation The no de simulation was run for a shorter p erio d of time b e cause of the diculty of running such a large simu lation Also the and no de simulations had a slightly larger session generation interval than the and no des networks in order to keep the to
et als pro duces b et they also used of the two pa rameters may account for the slightly decreased go o d put that their AODV simulations pro duced Another signicant dierence b etween their simulation and ours is that they set ROUTE DISCOVERY TIMEOUT to msec whereas we found the optimal value to b e
hello loss The combination
allowed
rreq retries
msec
To show that AODV nds
routes in a timely man ner we examined the route acquisition latency The route acquisition latency was computed by noting the simulation time when an initial RREQ was broadcast for a given destination and then noting the time when the rst RREP was received at the source For suc cessive RREQ retries for the same route the start time for the route was held at the time at which the
We note networks Reasons decreased go o dput ratio are a much greater collision rate due to the increase in the numb er of
tal numb er of sessions more manageable
that the results of the
and no de would have desired
are not for the
as go o d as we
Route Acquisition Latency (msec)
Route Acquisition Latency (msec)
of No des
Go o dput Ratio at sim end
Go o dput Ratio avg
Bandwidth Overhead Ratio
Avg Rte Acq Latency msec
Avg Path Length hops
Loss to Collision
Ro om Size m
x
x
x
x
Simulation Length sec
Generated Sessions
Completed Sessions
Ab orted Sessions
Table 3. Summary of S DATA Results
nodes and the longer paths causing a greater like lihood for collisions during the hopbyhop forward ing of the message and the added interference of all the hello messages Also the route acquisition la tency increased due to the larger average path length
and the
mission
access
mance
scalable adho c routing
networks this large we are pushing the
bilities of mobile networks as we are not aware of any other attempts to mo del networks of such a large size
We also ran simulations of the and no de net
of
same amount of control overhead ie the same numb er of RREQs and RREPs the voice sessions send many more data bits b ecause of the increased data packet size
additional
b ecause of increased However regardless
delay in control message trans comp etition for channel
net Draft submitted
group There are a numb er of
which may supp ort larger p opulations of adho c users
of the values AODV is currently
decreased p erfor one of the most
proto cols We
feel that with current capa
Inter
typ e describ ed in Sec typ e to stress the abil these simulations to results from the S DATA Table The two im portant results from these simulations are the good put ratio and the bandwidth overhead ratio The go o dput ratio for the voice session typ e was lower than that of the S DATA sessions This is due to the fact that there were signicantly more collisions due to the longer data packet lengths Also b e cause the data packets were larger and to ok longer to transmit we found that the queues of the no des frequently backed up b ecause they had to wait for channel access for p ossibly lengthy p erio ds of time causing delays in sending RREQs and RREPs On the other hand if we compare the bandwidth over head ratio b etween the two session typ es we nd that the voice sessions had more optimal results than the S DATA sessions This is b ecause for virtually the
works using the voice session tion We used this session
ities of AODV The results gether with the comparable session typ e are given in
Multicast as a
tions must b e considered when
gorithms for adho c networks
hanced AODV to provide multicast capability Mul ticast using AODV follows directly from the Route RequestRoute Reply message cycle and requires only one additional message typ e the Multicast Validation Message No des in the network that are memb ers of the same multicast group together with the nodes used as routers to connect group members form a bi directional multicast tree across which multicast data packets are relayed The MACT message is used to select the no de which a source no de cho oses as its next hop for the multicast tree Additionally there is a mul ticast group leader that is resp onsible for incrementing the multicast group sequence numb er More details of the multicast p ortion of AODV can b e found in
Current
Status and
Future Work
Currently AODV has b een
sp ecied in an
IETF manet working
or improve resp onse time to route queries or
increase
the capabilities of the
4.1. Multicast
proto col
to the
further improvements
basic
to ol
for conferencing designing routing al
applica We have already en
of No des
Session Typ e
S DATA
Voice
S DATA
Voice
Go o dput Ratio at sim end
Bandwidth Eciency
Avg Rte Acq Latency msec
Generated Sessions
4.2. Intermediate Node Route Rebuilding
Route rebuilding after a link breakage is currently the resp onsibility of the source no de However one al ternative to this metho d is to allow the no de upstream of the break to try to repair an active ie recently used route before sending the link failure notication Because the next hop with which it lost contact is likely to still b e in the near vicinity and have a valid route to the destination the TTL value of the RREQ sent by the intermediate node can be small so that the link failure can b e lo calized
A tradeo b etween quickly reestablishing the route and preventing the source no de from continuing to send data packets exists when allowing intermediate no des to rebuild routes Allowing intermediate no de route re building could provide for quicker route reconstruction and fewer dropp ed packets if the route is able to b e reconstructed quickly On the other hand more data packets will have b een lost during an unsuccessful re construction attempt than would have b een if a link failure notication had b een sent at the initial discov ery of the broken link We plan to investigate which metho d is sup erior in terms of go o dput and latency
4.3. Elimination of Hello Messages
Hello messages while allowing no des to learn ab out neighb or changes in a timely manner create extra con trol overhead and increase bandwidth consumption We chose to include hello messages in the design of AODV b ecause we did not want AODV to have to rely on an underlying MACsublayer proto col However we are currently investigating ways of eliminating the
This may for instance lead to transmitting additional route information along a backb one Such backb one no des might p erform intermediate varieties of route request propagation b efore relaying such requests in discriminately further improving bandwidth utiliza tion Another alternative could b e that mobile no des that are currently corresp onding might oer to ex change their lo cal routing tables with each other thus reducing further the setup time required for any of their mutual neighb ors to communicate with each other
QoS is another imp ortant feature of routing proto cols AODV has b een enhanced to provide basic QoS services namely delay and bandwidth assurances We plan to investigate the ecacy of these additions in the near future
need for op erate col
hello messages while still allowing AODV to indep endently from such an underlying proto
4.4. Locality of Association and QoS
We exp ect improvements to the latency of estab lishing routes by exploiting lo cality of asso ciation
Table 4. Comparison of Voice and S DATA Simulations
Conclusion
In summary we have presented a distance vector al gorithm that is suitable for use with adho c networks AODV avoids problems with previous prop osals no tably DSDV and has the following features
Nodes store only the routes that are needed
Need for broadcast is minimized
Reduces memory requirements and needless dupli
cations
Quick resp onse to link breakage
Lo opfree routes maintained by use of destination
sequence numb ers
Scalable to large p opulations of no des
Compared to DSDV and other algorithms which store continuously up dated routes to all destinations in the adho c network our algorithm has longer latency for route establishment but we have taken the following steps to alleviate this problem
A route to a destination may b e returned by any intermediate no de
Link breakages are rep orted immediately and routes are quickly reestablished
in active routes
Inactive routes are quickly aged out of the system b ecause they are more likely to go stale
Some improvements are enabled by careful b o okkeep
M Gerla and JC Tsai Multicluster Mobile Mul timedia Radio Network ACM J Wireless Networks July
S Guha and S Khuller Approximation Algorithms for Connected Dominating Sets University of Mary land College Park Technical Rep ort June
J Ioannidis and G Q M Jr The Design and Im plementation of a Mobile Internetworking Architec ture In Proceedings of the Winter USENIX Confer ence pages Jan
D Johnson and D Maltz Dynamic source routing in adho c wireless networks In Computer Communica
ing and by asso ciating each route with a neighb ors
We conclude
worstcase route
by the network diameter AODV is an excellent choice for adho c network establishment It will b e useful in applications for emergency services conferncing battleeld communications and communitybased net working We lo ok forward to further development of the proto col for quality of service intermediate route rebuilding and various interconnection top ologies with xed networks and the Internet
Acknowledgment
J Jubin and J D Tornow The DARPA Packet Ra
dio Network Proto cols In Proceedings of the IEEE
volume pages Jan
B M Leiner D L Nielson and F A Tobagi Issues
in Packet Radio Network Design Proceedings of the IEEE Special issue on Packet Radio Networks
J Macker and S C chairs Mobile Adho c Networks manet httpwwwietforghtmlchartersmanet charterhtml
S Murthy and J J GarciaLunaAceves A Rout ing Proto col for Packet Radio Networks In st ACM Intl Conference on Mobile Computing and Network ing Mobicom pages
A Parekh Selecting Routers in AdHoc Wireless Net work In Proceedings SBTIEEE Intl Telecommunica tions Symposium pages Aug
V D Park and M S Corson A Highly Adaptive Dis tributed Routing Algorithm for Mobile Wireless Net works In Proceedings of IEEE Conference on Computer Communications Infocom Apr
V D Park and M S Corson A Performance Compar ison of the Temp orallyOrdered Routing Algorithm and Ideal LinkState Routing In Proceedings of IEEE International Symposium on Systems and Communi cations IEEE Computer So ciety Press June
C Perkins and P Bhagwat Routing over Multi
that
establishment latency as
within the limits
Pravin Bhagwat was an early collab orator on the de sign of AODV and was resp onsible for a great deal of whiteb oard illumination and appropriate course correc tions George Aggelou participated in the implementa tion of some previous versions of the AODV simulation
and asked the right large p ercent of the tions
References
A L
questions Steve Fullmer wrote a
co de
that was used
for the simula
Alwan R
Kleinro ck J Short and J
Bagro dia
Commu R Bagrodia and W Liao Maisie A language for
mobile multimedia network IEEE Personal
nications June
design of ecient dicrete event simulation networks IEEE Transactions on Software ing Apr
computer Engineer
J Broch D A Maltz D B Johnson YC Hu and J Jetcheva A Performance Comparison of MultiHop Wireless AdHo c Network Routing Proto cols In In Pro ceedings of the Fourth Annual ACMIEEE Inter national Conference on Mobile Computing and Net working Oct
M S Corson and A Ephremides A Distributed Rout ing Algorithm for Mobile Wireless Networks ACM J Wireless Networks Jan
S Corson J Macker and S Batsell Architectural Considerations
for Mobile Mesh Networking work in progress httptonnantitdnrlnav ymil mmnetmmn etRFCtxt
B Das and V Bharghavan Routing in Adho c Net works Using Minimum Connected Dominating Sets In IEEE International Conference on Communications ICC June
hop Wireless Network of COMM Computer Oct
C E Perkins and E M Royer Ad Distance Vector AODV Routing
ao dvtxt Nov work in progress
N Bamb os Villasenor
M Gerla Adaptive
list of active imp osed by
determined
tions Review
Proceedings of SIGCOMM Aug
C Perkins Editor IP Oct
A S Tanenbaum Computer Networks rd Edition chapter pages Prentice Hall Englewo o d Clis edition
F Teraoka Y Yokote and M Tokoro A Network Ar chitecture Providing Host Migration Transparency In Proceedings of the SIGCOMM Conference Com munications Architectures Protocols pages Sept
Mobile Computers SIG
Communications
Review
Mobility Supp ort RFC
Ho c On draftietfmanet
Demand
A Pro of of the Lo opfree Prop erty
Verifying that AODV establishes only lo opfree routes is easy because of the eect of the destination se quence numb er on the maintenance of routes Supp ose that there is a loop in a route to a destination Z and that no des Xi are the no des in the lo op for i n As a matter of terminology we say that Xi points to Xj symbolically Xi X j if the routing table entry of Xi for destination Z shows no de Xj as the next hop
equality would have to hold the instant the lo op was
to Z Then for each Xi Xi X i for and furthermore Xn X
i n
Letting k one could
Let Ti b e the destination sequence
the route entry at Xi for destination Z Then Ti Ti whenever Xi Xi b ecause of the pro cessing of the RREP sp ecied in Section Since T T Tn T evidently the destination sequence numb ers are the same for
every if it
no de were
Xi in the routing
lo op routing
p ossible
to create a
numb er
Moreover lo op this
for
created Furthermore
quence numb ers are all the same the next hop infor
mation must have b een
the same RREP transmitted by the destination Z
b ecause the destination se
derived at every no de Xi from Consider now the metrics mi to the destination Z
Since Xi Xi only if mi mi mn n But because Xn X mn and n this is a contradiction
then m m
An b e the always
just by
point to X instead of of X This works
inductive argument is also p ossible
minimum length of a
construct another routing lo op of length
routing lo op
mo difying the routing table at the no de Xk to
k showing that it is
a routing lo op of length
using destination sequence
shows that routing lo ops cannot have length
routing lo ops cannot contain more than one no de But this means that there cannot b e any routing lo ops
only p ossible to Then a simple
numb ers
RREP
handling so that
k
as long as construct argument