Abstract

An ad hoc mobile network is a collection of mobile nodes that are dynamically and arbitrarily located in such a manner that the interconnections
between nodes are capable of changing on a continual basis. In order to facilitate communication within the network, a routing protocol is used
to discover routes between nodes. The primary goal of such an ad hoc network routing protocol is correct and efficient route establishment
between a pair of nodes so that messages may be delivered in a timely manner. Route construction should be done with a minimum of overhead
and bandwidth consumption. This article examines routing protocols for ad hoc networks and evaluates these protocols based on a given set of
parameters. The: article provides an overview of eight differcnt protocols by presenting their characteristics and functionality,and then provides a
comparison and discussion of their respective merits and drawbacks.

A Review of Current Routing Protocols for

Ad Hoc Mobile Wireless Networks
ELIZABETHM . ROYER, UNIVERSITY OF C A L I F O R N I A BARBARA
SANTA
,
CHAI-KEONG TOH, G E O R G I AINSTITUTE OF TECHNOLOGY

ince their emergence in the

1970s, wireless networks have become increasingly popular in
Existing Ad Hoc Routing Protocols
the computing industry. This is particularly true within the Since the advent of Defense Advanced Research Projects
past decade, which has seen wireless networks being adapted Agency (DARPA) packet radio networks in the early 1970s
to enable mobility. There are currently two variations of [l],numerous protocols have been developed for ad hoc
mobile wireless networks. The first is known as the infrastruc- mobile networks. Such protocols must deal with the typical
tured network (i.e., a network with fixed and wired gateways). limitations of these networks, which include high power
The bridges for these networks are known as base stations. A consumption, low bandwidth, and high e r r o r rates. As
mobile unit within these networks connects to, and communi- shown in Fig. 1, these routing protocols may generally be
cates with, the nearest base station that is within its communi- categorized as:
cation radius. As the mobile travels out of range of one base Table-driven
station and into the range of another, a “handoff’ occurs from Source-initiated (demand-driven)
the old base station to the new, and the mobile is able to con- Solid lines in this figure represent direct descendants, while
tinue communication seamlessly throughout the network. Typ- dotted lines depict logical descendants. Despite being designed
ical applications of this type of network include office wireless for the same type of underlying network, the characteristics of
local area networks (WLANs). each of these protocols are quite distinct. The following sec-
The second type of mobile wireless network is the infras- tions describe the protocols and categorize them according to
tructureless mobile network, commonly known as an ad hoc their characteristics.
network. Infrastructureless networks have no fixed routers; all
nodes are capable of movement and can be connected dynami- Table-Driven Routina ProtocoIs
J
cally in an arbitrary manner. Nodes of these networks function Table-driven routing protocols attempt to maintain consistent,
as routers which discover and maintain routes to other nodes up-to-date routing information from each node to every other
in the network. Example applications of ad hoc networks are node in the network. These protocols require each node to
emergency search-and-rescue operations, meetings or conven- maintain one or more tables to store routing information, and
tions in which persons wish to quickly share information, and they respond to changes in network topology by propagating
data acquisition operations in inhospitable terrain. updates throughout the network in order to maintain a consis-
This article examines routing protocols designed for these tent network view. The areas in which they differ are the
ad hoc networks by first describing the operation of each of number of necessary routing-related tables and the methods
the protocols and then comparing their various characteris- by which changes in network structure are broadcast. The fol-
tics. The remainder of the article is organized as follows. lowing sections discuss some of the existing table-driven ad
The next section presents a discussion of two subdivisions of hoc routing protocols.
ad hoc routing protocols. Another section discusses current
table-driven protocols, while a later section describes those Destination-Sequenced Distance-Vector Routing - The Des-
protocols which are classified as on-demand. The article tination-Sequenced Distance-Vector Routing protocol
then presents qualitative comparisons of table-driven proto- (DSDV) described in [2] is a table-driven algorithm based on
cols, followed by demand-driven protocols, and finally a gen- the classical Bellman-Ford routing mechanism [ 3 ] . The
eral comparison of table-driven and on-demand protocols. improvements made to the Bellman-Ford algorithm include
Applications and challenges facing ad hoc mobile wireless freedom from loops in routing tables.
networks are discussed; and finally, the last section con- Every mobile node in the network maintains a routing
cludes the article. table in which all of the possible destinations within the net-

46 1070-9916/99/$10.00 0 1999 IEEE IEEE Personal Communications * April 1999

work and the number of hops to each destination are record-
ed. Each entry is marked with a sequence number assigned by
the destination node. T h e sequence numbers enable the
mobile nodes to distinguish stale routes from new ones, there-
by avoiding the formation of routing loops. Routing table
updates are periodically transmitted throughout the network
in order to maintain table consistency. T o help alleviate the
potentially large amount of network traffic that such updates
can generate, route updates can employ two possible types of
packets. The first is known as afull dump. This type of packet
carries all available routing information and can require mul- W Figure 1. Categorization of ad hoc routingprotocols.
tiple network protocol data units (NPDUs). During periods of
occasional movement, these packets are transmitted infre-
quently. Smaller incremental packets are used to relay only rithm within the cluster. The disadvantage of having a clus-
that information which has changed since the last full dump. ter head scheme is that frequent cluster head changes can
Each of these broadcasts should fit into a standard-size adversely affect routing protocol performance since nodes
NPDU, thereby decreasing the amount of traffic generated. are busy in cluster head selection rather than packet relay-
The mobile nodes maintain an additional table where they ing. Hence, instead of invoking cluster head reselection
store the data sent in the incremental routing information every time the cluster membership changes, a Least Cluster
packets. Change (LCC) clustering algorithm is introduced. Using
New route broadcasts contain the address of the destina- LCC, cluster heads only change when two cluster heads
tion, t h e n u m b e r of hops to reach t h e destination, t h e come into contact, or when a node moves out of contact of
sequence number of the information received regarding the all other cluster heads.
destination, as well as a new sequence number unique to the CGSR uses DSDV as the underlying routing scheme, and
broadcast [ 2 ] . T h e r o u t e labeled with t h e most r e c e n t hence has much of the same overhead as DSDV. However, it
sequence number is always used. I n t h e event t h a t two modifies DSDV by using a hierarchical cluster-head-to-gate-
updates have the same sequence number, the route with the way routing approach to route traffic from source to destina-
smaller metric is used in order to optimize (shorten) the tion. Gateway nodes are nodes that are within communication
path. Mobiles also keep track of the settling time of routes, range of two or more cluster heads. A packet sent by a node
or the weighted average time that routes to a destination will is first routed to its cluster head, and then the packet is rout-
fluctuate before the route with the best metric is received ed from the cluster head to a gateway to another cluster head,
(see [ 2 ] ) .By delaying the broadcast of a routing update by and so on until the cluster head of the destination node is
the length of the settling time, mobiles can reduce network reached. The packet is then transmitted to the destination.
traffic and optimize routes by eliminating those broadcasts Figure 2 illustrates an example of this routing scheme. Using
that would occur if a better route was discovered in the very this method, each node must keep a “cluster member table”
near future. where it stores the destination cluster head for each mobile
node in the network. These cluster member tables are broad-
Clusterhead Gateway Switch Routing - The Clusterhead cast by each node periodically using the DSDV algorithm.
Gateway Switch Routing (CGSR) protocol differs from the Nodes update their cluster member tables on reception of
previous protocol in the type of addressing and network such a table from a neighbor.
organization scheme employed. Instead of a “flat” network, I n addition to t h e cluster member table, e a c h n o d e
CGSR is a clustered multihop mobile wireless network with must also maintain a routing table which is used to deter-
several heuristic routing schemes [4]. The authors state that mine the next hop in order to reach the destination. O n
by having a cluster head controlling a g r o u p of ad hoc receiving a packet, a node will consult its cluster member
nodes, a framework for code separation (among clusters), table and routing table to determine the nearest cluster
channel access, routing, and bandwidth allocation can be head along the route to the destination. Next, the node
achieved. A cluster head selection algorithm is utilized to will check its routing table to determine the next hop used
elect a node as the cluster head using a distributed algo- to reach the selected cluster head. It then transmits the
packet to this node.

The Wireless Routing Protocol - The Wireless Routing Pro-

tocol (WRP) described in [ 5 ] is a table-based protocol with
the goal of maintaining routing information among all nodes
in the network. Each node in the network is responsible for
maintaining four tables:
Distance table
Routing table
Link-cost table
Message retransmission list (MRL) table
Each entry of the MRL contains the sequence number of the
update message, a retransmission counter, an acknowledg-
ment-required flag vector with one entry per neighbor, and a
list of updates sent in the update message. The MRL records
which updates in an update message need to be retransmit-
ted and which neighbors should acknowledge the retransmis-
sion [5].
H Figure 2. CGSR: routingfrom node 1 to node 8. Mobiles inform each other of link changes through the use

IEEE Personal Communications April 1999 47

of update messages. An update message is sent only between tain the most recent route information. Each node maintains
neighboring nodes and contains a list of updates (the destina- its own sequence number, as well as a broadcast ID. The
tion, the distance to the destination, and the predecessor of broadcast ID is incremented for every RREQ the node initi-
the destination), as well as a list of responses indicating ates, and together with the node’s IP address, uniquely identi-
which mobiles should acknowledge (ACK) the update. fies an RREQ. Along with its own sequence number and the
Mobiles send update messages after processing updates from broadcast ID, the source node includes in the R R E Q the
neighbors or detecting a change in a link to a neighbor. In most recent sequence number it has for the destination. Inter-
the event of the loss of a link between two nodes, the nodes mediate nodes can reply to the R R E Q only if they have a
send update messages to their neighbors. The neighbors then route to the destination whose corresponding destination
modify their distance table entries and check for new possible sequence number is greater than or equal to that contained in
paths through other nodes. Any new paths are relayed back the RREQ.
to the original nodes so that they can update their tables During the process of forwarding the RREQ, intermediate
accordingly. nodes record in their route tables the address of the neighbor
Nodes learn of the existence of their neighbors from the from which the first copy of the broadcast packet is received,
receipt of acknowledgments and other messages. If a node is thereby establishing a reverse path. If additional copies of the
not sending messages, it must send a hello message within a same RREQ are later received, these packets are discarded.
specified time period to ensure connectivity. Otherwise, the Once the RREQ reaches the destination or an intermediate
lack of messages from the node indicates the failure of that node with a fresh enough route, the destinationiintermediate
link; this may cause a false alarm. When a mobile receives a node responds by unicasting a route reply (RREP) packet
hello message from a new node, that new node is added to the back to the neighbor from which it first received the RREQ
mobile’s routing table, and the mobile sends the new node a (Fig. 3b). As the RREP is routed back along the reverse path,
copy of its routing table information. nodes along this path set up forward route entries in their
Part of the novelty of WRP stems from the way in which route tables which point to the node from which the RREP
it achieves locip freedom. In WRP, routing nodes communi- came. These forward route entries indicate the active forward
cate the distance and second-to-last hop information for route. Associated with each route entry is a route timer which
each destination in the wireless networks. WRP belongs to will cause the deletion of the entry if it is not used within the
the class of path-finding algorithms with an important excep- specified lifetime. Because the RREP is forwarded along the
tion. It avoids the “count-to-infinity” problem [6] by forcing path established by the RREQ, AODV only supports the use
each node to perform consistency checks of predecessor of symmetric links.
information reported by all its neighbors. This ultimately Routes are maintained as follows. If a source node moves,
(although not instantaneously) eliminates looping situations it is able to reinitiate the route discovery protocol to find a
and provides faster route convergence when a link failure new route to the destination. If a node along the route moves,
event occurs. its upstream neighbor notices the move and propagates a link
failure notification message (an RREP with infinite metric) to
Source-Initiated On-Demand Routing each of its active upstream neighbors to inform them of the
A different approach from table-driven routing is source-initi- erasure of that part of the route [ 7 ] .These nodes in turn
ated on-demand routing. This type of routing creates routes
only when desired by the source node. When a node requires
a route to a destination, it initiates a route discovery process
within the network. This process is completed once a route is
found or all possible route permutations have been examined. Destination
Once a route has been established, it is maintained by a route
maintenance procedure until either the destination becomes
inaccessible along every path from the source or until the Source
route is no longer desired.

Ad Hoc On-Demand Distance Vector Routing - The Ad Hoc

On-Demand Distance Vector (AODV) routing protocol
described in [ 7 ]builds on the DSDV algorithm previously
U
described. AODV is an improvement on DSDV because it
typically minimizes the number of required broadcasts by cre- (a) Propagation of the RREQ
ating routes on a demand basis, as opposed to maintaining a
complete list of routes as in the DSDV algorithm. The authors
of AODV classify it as apure on-demand route acquisition sys-
tem, since nodes that are not on a selected path do not main-
tain routing information o r participate in routing table
exchanges [7].
/m-p Destination

When a source node desires to send a message to some

Source
destination node and does not already have a valid route to
that destination, it initiates a path discovery process to locate
the other node. It broadcasts a route request (RREQ) packet
to its neighbors, which then forward the request to their
neighbors, and so on, until either the destination or an inter-
mediate node with a “fresh enough” route to the destination Ly

is located. Figure 3a illustrates the propagation of the broad- (b) Path of the RREP to the source
cast RREQs across the network. AODV utilizes destination
sequence numbers to ensure all routes are loop-free and con- IFigure 3. AODVroute discovety.

48 IEEE Personal Communications April 1999

IU Figure 4. Creation of the route record in DSR.

propagate the link failure notification to their upstream neigh- is the destination, it places the route record contained in the
bors, and so on until the source node is reached. The source route request into the route reply. If the responding node is
node may then choose to reinitiate route discovery for that an intermediate node, it will append its cached route to the
destination if a route is still desired. route record and then generate the route reply. To return the
An additional aspect of the protocol is the use of hello route reply, the responding node must have a route to the
messages, periodic local broadcasts by a node to inform each initiator. If it has a route to the initiator in its route cache, it
mobile node of other nodes in its neighborhood. Hello mes- may use that route. Otherwise, if symmetric links are sup-
sages can be used to maintain the local connectivity of a ported, the node may reverse the route in the route record. If
node. However, the use of hello messages is not required. symmetric links are not supported, the node may initiate its
Nodes listen for retransmission of data packets to ensure that own route discovery and piggyback the route reply on the
the next hop is still within reach. If such a retransmission is new route request. Figure 4b shows the transmission of the
not heard, the node may use any one of a number of tech- route reply with its associated route record back t o the
niques, including the reception of hello messages, to deter- source node.
mine whether the next hop is within communication range. Route maintenance is accomplished through the use of
The hello messages may list the other nodes from which a route error packets and acknowledgments. Route error packets
mobile has heard, thereby yielding greater knowledge of net- are generated at a node when the data link layer encounters a
work connectivity. fatal transmission problem. When a route error packet is
received, the hop in error is removed from the node’s route
Dynamic Source Routing - The Dynamic Source Routing cache and all routes containing the hop are truncated at that
(DSR) protocol presented in [8] is an on-demand routing point. In addition to route error messages, acknowledgments
protocol that is based on the concept of source routing. are used to verify the correct operation of the route links.
Mobile nodes are required to maintain route caches that Such acknowledgments include passive acknowledgments,
contain the source routes of which the mobile is aware. where a mobile is able to hear the next hop forwarding the
Entries in the route cache are continually updated as new packet along the route.
routes are learned.
The protocol consists of two major phases: route discovery TemporaZly Ordered Routing Algorithm - The Temporally
and route maintenance. When a mobile node has a packet t o Ordered Routing Algorithm (TORA) is a highly adaptive
send to some destination, it first consults its route cache to loop-free distributed routing algorithm based on the concept
determine whether it already has a route to the destination. If of link reversal [lo]. TORA is proposed to operate in a high-
it has an unexpired route to the destination, it will use this ly dynamic mobile networking environment. It is source-initi-
route to send the packet. On the other hand, if the node does ated and provides multiple routes for any desired
not have such a route, it initiates route discovery by broad- source/destination pair. The key design concept of TORA is
casting a route request packet. This route request contains the the localization of control messages to a very small set of
address of the destination, along with the source node’s nodes near the occurrence of a topological change. To
address and a unique identification number. Each node accomplish this, nodes need to maintain routing information
receiving the packet checks whether it knows of a route to the about adjacent (one-hop) nodes. The protocol performs three
destination. If it does not, it adds its own address to the route basic functions:
record of the packet and then forwards the packet along its
outgoing links. To limit the number of route requests propa-
gated on the outgoing links of a node, a mobile only forwards
- Route creation
Route maintenance
Route erasure
the route request if the request has not yet been seen by the During the route creation and maintenance phases,
mobile and if the mobile’s address does not already appear in nodes use a “height” metric to establish a directed acyclic
the route record. graph (DAG) rooted at the destination. Thereafter, links
A route reply is generated when the route request reaches are assigned a direction (upstream or downstream) based
either the destination itself, or an intermediate node which o n the relative height metric of neighboring nodes, as
contains in its route cache an unexpired route to the destina- shown in Fig. Sa. This process of establishing a D A G is
tion [9]. By the time the packet reaches either the destination similar to the query/reply process proposed in Lightweight
or such an intermediate node, it contains a route record Mobile Routing (LMR) [ l l ] . In times of node mobility the
yielding the sequence of hops taken. Figure 4a illustrates the DAG route is broken, and route maintenance is necessary
formation of the route record as the route request propagates to reestablish a DAG rooted at the same destination. As
through the network. If the node generating the route reply shown in Fig. 5b, upon failure of the last downstream link,

IEEE Personal Communications April 1999 49

a node generates a new reference level which results in the TORA’s route erasure phase essentially involves flooding a
propagation of that reference level by neighboring nodes, broadcast clear packet (CLR) throughout the network to
effectively coordinating a structured reaction to the failure. erase invalid routes.
Links are reversed to reflect the change in adapting to the In T O R A there is a potential for oscillations to occur,
new reference level. This has the same effect as reversing especially when multiple sets of coordinating nodes are con-
the direction of o n e o r m o r e links when a node has n o currently detecting partitions, erasing routes, and building
downstream links. new routes based on each other. Because TORA uses inter-
Timing is an important factor for T O R A because the nodal coordination, its instability problem is similar to the
“height” metric is dependent on the logical time of a link fail- “count-to-infinity’’problem in distance-vector routing proto-
ure; TORA assumes that all nodes have synchronized clocks cols, except that such oscillations are temporary and route
(accomplished via an external time source such as the Global convergence will ultimately occur.
Positioning System). TORA’s metric is a quintuple comprising
five elements. namely: Associativity-Based Routing - A totally different approach
Logical time of a link failure in mobile routing is proposed in [12]. T h e Associativity-
The unique ID of the node that defined the new reference Based Routing (ABR) protocol is free from loops, deadlock,
level and packet duplicates, and defines a new routing metric for
A reflection indicator bit ad hoc mobile networks. This metric is known as the degree
A propagation ordering parameter of association stability. In ABR, a route is selected based on
The unique ID of the node the degree of association stability of mobile nodes. Each
The first three elements collectively represent the refer- node periodically generates a beacon to signify its existence.
ence level. A new reference level is defined each time a When received by neighboring nodes, this beacon causes
node loses ilrs last downstream link due to a link failure. their associativity tables to be updated. F o r each beacon
received, the associativity tick of the current node with

r-
f
respect to the beaconing node is incremented. Association
stability is defined by connection stability of one node with
respect to another node over time and space. A high degree
of association stability may indicate a low state of node
mobility, while a low degree may indicate a high state of
node mobility. Associativity ticks are reset when the neigh-
bors of a node or the node itself move out of proximity. A
fundamental objective of ABR is to derive longer-lived
routes for ad hoc mobile networks.
The three phases of ABR are:
Route discovery
Route reconstruction (RRC)
Route deletion
The route discovery phase is accomplished by a broad-
c a s t q u e r y a n d a w a i t - r e p l y (BQ-REPLY) cycle. A n o d e
desiring a r o u t e broadcasts a BQ message in search of
mobiles that have a route to t h e destination. All nodes
receiving the query (that are not the destination) append
their addresses and their associativity ticks with their neigh-
bors along with QoS information to the query packet. A
successor node erases its upstream node neighbors’ associa-
tivity tick entries and retains only the entry concerned with
itself and its upstream node. I n this way, each resultant
packet arriving at the destination will contain the associativ-
ity ticks of the nodes along the route to the destination. The
destination is then able to select the best route by examin-
ing the associativity ticks along each of the paths. When
multiple paths have the same overall degree of association
stability, the route with the minimum number of hops is
selected. The destination then sends a REPLY packet back
t o t h e s o u r c e along this p a t h . N o d e s p r o p a g a t i n g t h e
REPLY mark their routes as valid. All other routes remain
inactive, and the possibility of duplicate packets arriving at
the destination is avoided.
RRC may consist of partial route discovery, invalid route
erasure, valid route updates, and new route discovery, depend-
ing on which node(s) along the route move. Movement by the
source results in a new BQ-REPLY process, as shown in Fig.
6a. The RN[l] message is a route notification used to erase
the route entries associated with downstream nodes. When
the destination moves, the immediate upstream node erases
IFigure 5. a ) Route creation(showing link direction assign- its route and determines if the node is still reachable by a
ment); b) route maintenance (showing the link reversal phe- localized query (LQ[HI) process, where H refers to the hop
nomenon) in T O M . count from the upstream node to the destination (Fig. 6b). If

50 IEEE Personal Communications April 1999

the destination receives the LQ packet, it
REPLYs with the best partial route; other-
wise, the initiating node times out and the
process backtracks to the next upstream
node. Here an RN[O] message is sent to the
next upstream node to erase the invalid
route and inform this node that it should
invoke the LQ[H] process. If this process
results in backtracking more than halfway to
the source, the LQ process is discontinued
and a new BQ process is initiated at the
source. W Figure 6.Route maintenancefor source and destination movement in ABR.
When a discovered route is n o longer
desired, the source node initiates a route
delete ( R D ) broadcast so that all nodes along t h e route The DRP is responsible for the maintenance of the Sig-
update their routing tables. The RD message is propagated by nal Stability Table (SST) and Routing Table (RT). T h e
a full broadcast, as opposed to a directed broadcast, because SST records t h e signal strength of neighboring nodes,
the source node may not be aware of any route node changes which is obtained by periodic beacons from the link layer
that occurred during RRCs. of each neighboring node. Signal strength may be recorded
as either a strong or weak channel. All transmissions are
Signal Stability Routing - Another on-demand protocol is received by, and processed in, the DRP. After updating all
the Signal Stability-Based Adaptive Routing protocol (SSR) appropriate table entries, the DRP passes a received packet
presented in [13]. Unlike the algorithms described so far, SSR to the SRP.
selects routes based on the signal strength between nodes and The SRP processes packets by passing the packet up the
a node’s location stability. This route selection criteria has the stack if it is the intended receiver or looking up the destina-
effect of choosing routes that have “stronger” connectivity. tion in the RT and then forwarding the packet if it is not. If
SSR can be divided into two cooperative protocols: t h e no entry is found in the RT for the destination, a route-search
Dynamic Routing Protocol (DRP) and the Static Routing process is initiated to find a route. Route requests are propa-
Protocol (SRP). gated throughout the network, but are only forwarded to the
next hoD if thev are received over
strong ihannels and have not been
previoudy proccsscd (to prcvcnt
looping). Thc dcstination chootcs
I Time complexity (link addition/failure) I thefirst’arriving route-search pack-
et to send back because it is most
Communication compiexity (link additiordfailure) O h = N) = PI) O(x = AI) probable that the packet arrived
over the shortest and/or least con-
I Routing philosophy I Flat IHierarchical ]Flat* I gested path. The DRP then revers-
Loop-free YeS Yes Yes, but not es the selected route and sends a
inrtantanews route-reply message back to the ini-
tiator. The DRP of the nodes along
I Multicast capability I No ]NO** INo I the path update their RTs accord-
ingly.
Number of required tables TWO Two Four

1 Frequency of update transmissions IPeriodicallyAPeriodically IPeriodically

and as need
I
and as needed
Route-search packets arriving at
the destination have necessarily
chosen the path of strongest signal
stability, since the p a c k e t s a r c
! Updates transmitted to M g h h Neighbors~~*hbo~ dropped at a node if they havc
I cluster head arrived over a weak channel. I f
I Utilizes seauence numbers IYes IYes lyes I t h e r e is no route-reply message
received at the source within a spe-
; Utilizes hello messages Yes No YeS cific timeout period, t h e source
changes t h e P R E F field in t h e
1 Critical nodes
INo l c i s t e r headdNo I header to indicate that weak chan-
nels are acceptable, since these may
b e t h e only links over which t h e
packet can be propagated.
Abbreviations: When a failed link is detected
N = Number of nodes in the network within the network, the intermedi-
d = Network diameter ate nodes send an error message to
h = Height of routing tree the source indicating which channel
x = Number of nodes affected by a topological change has failed. The source then initiates
another route-search process to
* While WRP uses flat addressing, it can be used hierarchically [15]. find a new path to the destination.
** The protocol itself currently does not support multicast; however, there is a separate The Source sends an mes-
protocol described in [16], which runs on top of CGSR and provides multicast capability.
sage to notify all nodes of the bro-

IEEE Personal Communications April 1999 51

 Comparisons the underlying routing protocol. Routing in CGSR occurs
over cluster heads and gateways. A cluster head table is nec-
The following sections provide comparisons of the previously essary in addition to the routing table. O n e advantage of
described routing algorithms. The next section compares CGSR is that several heuristic methods can be employed to
table-driven protocols, and another section compares on- improve the protocol’s performance. These methods include
demand protocols. A later section presents a discussion of priority token scheduling, gateway code scheduling, and
the two classes of algorithms. In Tables 1 and 2, time com- path reservation [4].
plexity is defined as the number of steps needed to perform a The W R P protocol differs from the other protocols in
protocol operation, and communication complexity is the several ways. W R P requires each node to maintain four
number of messages needed to perform a protocol operation routing tables. This can lead to substantial memory require-
[ll, 141. Also, the values for these metrics represent worst- ments, especially when the number of nodes in the network
case behavior. is large. Furthermore, the W R P protocol requires the use
of hello packets whenever there are no recent packet trans-
Table-Driven Protocols missions from a given node. T h e hello packets consume
The discussion here is based on Table 1. As stated earlier, bandwidth and disallow a node to enter sleep mode. How-
DSDV routing is essentially a modification of the basic Bell- ever, although it belongs to the class of path-finding algo-
man-Ford routing algorithm. The modifications include the rithms, W R P has an advantage over o t h e r path-finding
guarantee of loop-free routes and a simple route update algorithms because it avoids the problem of creating tem-
protocol. While only providing one path to any given desti- porary routing loops that these algorithms have through the
nation, DSDV selects the shortest path based on the num- verification of predecessor information, as described in an
ber of hops to the destination. DSDV provides two types of earlier section.
update messages, one of which is significantly smaller than Having discussed the operation and characteristics of
the other. T h e smaller u p d a t e message can b e used for each of the existing table-driven routing protocols, it is
incremental updates so that the entire routing table need important to highlight the differences. During link failures,
not be transmitted for every change in network topology. W R P has lower time complexity than DSDV since it only
However, DSDV is inefficient because of the requirement informs neighboring nodes about link status changes. Dur-
of periodic update transmissions, regardless of the number ing link additions, hello messages are used as a presence
of changes in the network topology. This effectively limits indicator such that the routing table entry can be updated.
the number of nodes that can connect to the network since Again, this only affects neighboring nodes. I n C G S R ,
the overhead grows as O ( n 2 ) .In CGSR, DSDV is used as because routing performance is dependent on the status of

1 Communication complexity (initialization) IO(2N) I O(2N) I O(2N) I OW + Y ) I O(N + Y ) I

I
Communication complexity (postfailure) O(2N) O(zN) O(W

I Routing philosophy I Flat I Flat I Flat I Flat I Flat I

Loop-free Yes YeS Yes Yes Yes

I Multicast capability IYes I No I NO** I No IN0 I

Beaconing requirements No No No Yes Yes

I Multiple route possibilities I No I Yes I Yes I No IN0 I

Routes maintained in Route table Route cache Route table Route table Route table
I
1 Utilizes route cachdtable expiration timers lYes I No I No I No I No I
Route reconfiguration methodology Erase route; Erase route; Link reversal; Localized Erase route;
notify source notify source route repair broadcast query notify source
Routing metric

1 Freshest and
shortest path
1 Shortest path

1 Shortest path

1 1
Associativity and Associativity and
shortest path and stability
others*** 1

W Table 2. Comparisons of the characteristics of source-initiated on-demand ad hoc routingprotocols.

52 IEEE Personal Communications April 1999

specific nodes (cluster head, gateway, o r normal nodes), of the novelty of TORA stems from its creation of DAGS to
time complexity of a link failure associated with a cluster aid route establishment. One of the advantages of TORA is
head is higher than DSDV, given the additional time need- its support for multiple routes. TORA and DSR are the only
ed t o perform cluster head reselection. Similarly, this on-demand protocols considered here which retain multiple
applies to the case of link additions associated with the clus- route possibilities for a single source/destination pair. Route
ter head. There is no gateway selection in CGSR since each reconstruction is not necessary until all known routes to a
node declares it is a gateway node to its neighbors if it is destination are deemed invalid, and hence bandwidth can
responding t o multiple radio codes. If a gateway node potentially be conserved because of the necessity for fewer
moves out of range, the routing protocol is responsible for route rebuildings. Another advantage of TORA is its sup-
routing the packet to another gateway. port for multicast. Although, unlike AODV, TORA does not
In terms of communication complexity, since DSDV, incorporate multicast into its basic operation, it functions as
CGSR, and WRP use distance vector shortest-path routing as the underlying protocol for the Lightweight Adaptive Multi-
the underlying routing protocol, they all have the same degree cast Algorithm (LAM), and together the two protocols pro-
of complexity during link failures and additions. vide multicast capability [18]. TORA’s reliance on
synchronized clocks, while a novel idea, inherently limits its
Source-In itia ted On -Deman d Routing Protocols applicability. If a node does not have GPS or some other
Table 2 presents a comparison of AODV, DSR, TORA, external time source, it cannot use the algorithm. Addition-
ABR, and SSR. ally, if the external time source fails, the algorithm will cease
AODV employs a route discovery procedure similar to to operate. Furthermore, route rebuilding in TORA may not
DSR; however, there are a couple of important distinctions. occur as quickly as in the other algorithms due to the poten-
The most notable of these is that the overhead of DSR is tial for oscillations during this period. This can lead to
potentially larger than that of AODV since each DSR packet potentially lengthy delays while waiting for the new routes to
must carry full routing information, whereas in AODV pack- be determined.
ets need only contain the destination address. Similarly, the ABR is a compromise between broadcast and point-to-
route replies in DSR are larger because they contain the point routing, and uses the connection-oriented packet for-
address of every node along the route, whereas in AODV warding approach. Route selection is primarily based on the
route replies need only carry the destination IP address and aggregated associativity ticks of nodes along the path. Hence,
sequence number. Also, the memory overhead may be slight- although the resulting path does not necessarily result in the
ly greater in DSR because of the need to remember full smallest possible number of hops, the path tends to be longer-
routes, as opposed to only next hop information in AODV. A lived than other routes. A long-lived route requires fewer
further advantage of AODV is its support for multicast [17]. route reconstructions and therefore yields higher throughput.
None of the other algorithms considered in this article cur- Another benefit of ABR is that, like the other protocols, it is
rently incorporate multicast communication. On the down- guaranteed to be free of packet duplicates. The reason is that
side, AODV requires symmetric links between nodes, and only the best route is marked valid, while all other possible
hence cannot utilize routes with asymmetric links. In this routes remain passive. ABR, however, relies on the fact that
aspect DSR is superior, since it does not require the use of each node is beaconing periodically. The beaconing interval
such links and can utilize asymmetric links when symmetric must be short enough to accurately reflect the spatial, tempo-
links are not available. ral, and connectivity state of the mobile hosts. This beaconing
The DSR algorithm is intended for networks in which requirement may result in additional power consumption.
the mobiles move at moderate speed with respect to packet However, experimental results obtained in [19] reveal that the
transmission latency [8]. Assumptions the algorithm makes inclusion of periodic beaconing has a minute influence on the
for operation are that the network diameter is relatively overall battery power consumption. Unlike DSR, ABR does
small and that the mobile nodes can enable a promiscuous not utilize route caches.
receive mode, whereby every received packet is delivered to The SSR algorithm is a logical descendant of ABR. It uti-
the network driver software without filtering by destination lizes a new technique of selecting routes based on the signal
address. An advantage of DSR over some of the other on- strength and location stability of nodes along the path. As in
demand protocols is that DSR does not make use of period- ABR, while the paths selected by this algorithm are not neces-
ic routing advertisements, thereby saving bandwidth and sarily shortest in hop count, they do tend to be more stable
reducing power consumption. Hence, the protocol does not and longer-lived, resulting in fewer route reconstructions. One
incur any overhead when there are no changes in network of the major drawbacks of the SSR protocol is that, unlike in
topology. Additionally, DSR allows nodes to keep multiple AODV and DSR, intermediate nodes cannot reply to route
routes to a destination in their cache. Hence, when a link requests sent toward a destination; this results in potentially
on a route is broken, the source node can check its cache long delays before a route can be discovered. Additionally,
for another valid route. If such a route is found, route when a link failure occurs along a path, the route discovery
reconstruction does not need to be reinvoked. In this case, algorithm must be reinvoked from the source to find a new
r o u t e recovery is faster than in many of t h e o t h e r on- path to the destination. No attempt is made to use partial
demand protocols. However, if there are no additional route recovery (unlike ABR), that is, to allow intermediate
routes to the destination in the source node’s cache, route nodes to attempt to rebuild the route themselves. AODV and
discovery must be reinitiated, as in AODV, if the route is DSR also do not specify intermediate node rebuilding. While
still required. O n the other hand, because of the small this may lead to longer route reconstruction times since link
diameter assumption and the source routing requirement, failures cannot be resolved locally without the intervention of
DSR is not scalable to large networks. Furthermore, as pre- the source node, the attempt and failure of an intermediate
viously stated, the need to place the entire route in both node to rebuild a route will cause a longer delay than if the
route replies and data packets causes greater control over- source node had attempted the rebuilding as soon as the bro-
head than in AODV. ken link was noticed. Thus, it remains to be seen whether
TORA is a “link reversal” algorithm that is best suited intermediate node route rebuilding is more optimal than
for networks with large dense populations of nodes [lo]. Part source node route rebuilding.

IEEE Personal Communications April 1999 53

 Availability of routing
information
Available when needed
I
Always available regardless of
need
interact on a commonly shared
whiteboard. Ad hoc mobile com-
Routing philosophy Flat Mostly flat, except for CGSR munication is particularly useful in
relaying information (status, situa-
1 Periodic route updates I1 Not required IRequired I tion awareness, etc.) via data, video,
I
and/or voice from one rescue team
Use localized route discovery Inform other nodes to achieve
Coping with mobility
as in ABR and SSR a consistent routing table i member to another over a small
handheld o r wearable wireless
Signaling traffic generated
of active routes (as in ABR) I
Grows with increasing mobility Greater than that of on-
demand routing I device. Again, this applies to law
enforcement personnel as well.
Current challenges for ad hoc
Quality of service support Few can support QoS, although Mainly shortest path as the I wireless networks include:
---___.-.___
most support shortest path
.-__..___ .___ ..-
00s metric
Table 3. Overall comparisons of on-demand versus table-driven routingprotocols.
i *Multicast [21]
*QoS support
Power-aware routing [22]
Location-aided routing [23]
As mentioned above, multicast
Table-Driven vs. On-Demand Routing is desirable to support multiparty wireless communications.
As discussed earlier, the table-driven ad hoc routing approach Since the multicast tree is no longer static (i.e., its topology
is similar to 1 he connectionless approach of forwarding pack- is subject to change over time), the multicast routing proto-
ets, with no regard to when and how frequently such routes col must be able to cope with mobility, including multicast
are desired. It relies on an underlying routing table update membership dynamics (e.g., leave and join). In terms of
mechanism that involves the constant propagation of routing QoS, it is inadequate to consider QoS merely at the network
information. This is not the case, however, for on-demand level without considering the underlying media access con-
routing protocols. When a node using an on-demand protocol trol layer [24]. Again, given the problems associated with the
desires a route to a new destination, it will have to wait until dynamics of nodes, hidden terminals, and fluctuating link
such a route can be discovered. On the other hand, because characteristics, supporting end-to-end QoS is a nontrivial
routing information is constantly propagated and maintained issue that requires in-depth investigation. Currently, there is
in table-driven routing protocols, a route to every other node a trend toward an adaptive QoS approach instead of the
in the ad hoc network is always available, regardless of “plain” resource reservation method with hard QoS guaran-
whether or not it is needed. This feature, although useful for tees. Another important factor is the limited power supply in
datagram traffic, incurs substantial signaling traffic and power handheld devices, which can seriously prohibit packet for-
consumption. Since both bandwidth and battery power are warding in an ad hoc mobile environment. Hence, routing
scarce resources in mobile computers, this becomes a serious traffic based on nodes’ power metrics is one way to distin-
limitation. Table 3 lists some of the basic differences between guish routes that are more long-lived than others. Finally,
the two classes of algorithms. instead of using beaconing or broadcast search, location-
Another consideration is whether a flat or hierarchical aided routing uses positioning information to define associ-
addressing scheme should be used. All of the protocols con- ated regions so t h a t the routing is spatially oriented and
sidered here, except for CGSR, use a flat addressing scheme. limited. This is analogous to associativity-oriented and
In [20] a discussion of the two addressing schemes is present- restricted broadcast in ABR.
ed. While flat addressing may be less complicated and easier Current ad hoc routing approaches have introduced several
to use, there are doubts as to its scalability. new paradigms, such as exploiting user demand, and the use
of location, power, and association parameters. Adaptivity and
p p ica tions and Cha Zleng es
A 1 self-configuration are key features of these approaches. How-
ever, flexibility is also important. A flexible ad hoc routing
Akin to packet radio networks, ad hoc wireless networks have protocol could responsively invoke table-driven andlor on-
an important role to play in military applications. Soldiers demand approaches based on situations and communication
equipped with multimode mobile communicators can now requirements. The “toggle” between these two approaches
communicate in an ad hoc manner without the need for fixed may not be trivial since concerned nodes must be “in sync”
wireless base stations. In addition, small vehicular devices with the toggling. Coexistence of both approaches may also
equipped with audio sensors and cameras can be deployed at exist in spatially clustered ad hoc groups, with intracluster
targeted regions to collect important location and environ- employing the table-driven approach and intercluster employ-
mental information which will be communicated back to a ing the demand-driven approach or vice versa. Further work is
processing node via ad hoc mobile communications. Ship-to- necessary to investigate the feasibility and performance of
ship ad hoc mobile communication is also desirable since it hybrid ad hoc routing approaches. Lastly, in addition to the
provides alternate communication paths without reliance on above, further research in the areas of media access control,
ground- or space-based communication infrastructures. security, service discovery, and Internet protocol operability is
Commercial scenarios for ad hoc wireless networks include: required before the potential of ad hoc mobile networking can
Conferences/meetings/lectures be realized.
Emergency services
Law enforcement
People today attend meetings and conferences with their
Concl usion
laptops, palmtops, and notebooks. It is therefore attractive to In this article we provide descriptions of several routing
have instant network formation, in addition to file and infor- schemes proposed for ad hoc mobile networks. We also
mation sharing without the presence of fixed base stations and provide a classification of these schemes according to the
systems administrators. A presenter can multicast slides and routing strategy (i.e., table-driven and on-demand). We
audio to intended recipients. Attendees can ask questions and have presented a comparison of these two categories of

54 IEEE Personal Communications April 1999

routing protocols, highlighting their features, differences, [151S. Murthy and J. J. Garcia-Luna-Aceves, "Loop-Free Internet Routing
Using Hierarchical Routing Trees," Proc. lNFOCOM '97, Apr. 7-1 1, 1997.
and characteristics. Finally, we have identified possible [161 C.-C. Chiang, M. Gerla, and L. Zhang, "Adaptive Shared Tree Multicast in
applications and challenges facing ad hoc mobile wireless Mobile Wireless Networks," Pnx. GLOBECOM '98. Nov. 1998. pp. 1817-22.
networks. While it is not clear that any particular algorithm [171C. E. Perkins and E. M. Royer, "Ad Hoc On Demand Distance Vector
or class of algorithm is the best for all scenarios, each pro- (AODV) Routing," IETF Internet draft, draft-ietf-manet-aodv-O2,txt, Nov.
tocol has definite advantages and disadvantages, and is well 1998 (work in progress).
[I81L. Jiand M. S. Corson, "A Lightweight Adaptive Multicast Algorithm,"
suited for certain situations. The field of ad hoc mobile Proc. GLOBECOM '98. Nov. 1998. pp. 1036-42.
networks is rapidly growing and changing, and while there [191C.-K. Toh and George Lin, "Implementing Associativity-Based Routing
are still many challenges that need to be met, it is likely for Ad Hoc Mobile Wireless Networks," Unpublished article, Mar. 1998.
that such networks will see widespread use within the next 1201 D. Baker et al., "Flat vs. Hierarchical Network Control Architecture,"
ARO/DARPA Wksp. Mobile Ad-Hoc Networking; http://www.isr.umd.
few years. edu/CourseWorkshops/MANET/program.html, Mar. 1997.
[211 M. Gerla, C.-C. Chiang, and L. Zhang, "Tree Multicast Strategies in
A ckn a wledgm en ts Mobile, Multihop Wireless Networks," ACMIBaltzer Mobile Networks
and Apps. 1.. 1998.
We would like to thank C.-C. Chiang, J . J. Garcia-Luna- 1221 S. Singh, M. Woo, and C. S. Raghavendra, "Power-Aware Routing in
Aceves, David Maltz, Vincent Park, and Charles Perkins for Mobile Ad Hoc Networks," Proc. ACMll€€€ MOBlCOM '98,Oct. 1998.
their assistance in ensuring accurate descriptions of the opera- [231Y. B. KO and N. H. Vaidya, "Location-Aided Routing (IAR) in Mobile Ad
tion of their protocols. Hoc Networks," Proc. ACMll€E€ MOBlCOM '98,Oct. 1998.
[241C. R. Lin and M. Gerla, "MACNPR: An Asynchronous Multimedia Multi-
hop Wireless Network," Proc. /€E€ lNFOCOM '97,Mar. 1997.
References
[I] J. Jubin and J. Tornow, "The DARPA Packet Radio Network Protocols,"
Proc. /E€€, vol. 75, no. 1, 1987, pp, 21-32.
[2]C. E. Perkins and P. Bhagwat, "Highly Dynamic Destination-Sequenced Biographies
Distance-Vector Routing (DSDV) for Mobile Computers," Comp. Com- ELIZABAETH M. ROVER ([email protected]) received her M.S. degree
mun. Rev., Oct. 1994, pp. 234-44. in electrical and computer engineering from the University of California,
[3]L. R. Ford Jr. and D. R. Fulkerson, Flows in Networks, Princeton Univ. Santa Barbara in December 1997, and her B.S. degrees in computer science
Press, 1962. and applied mathematics from Florida State University in April 1996. She is
[41C.-C. Chiang, "Routing in Clustered Multihop, Mobile Wireless Networks with currently a Ph.D. student in the Electrical and Computer Engineering
Fading Channel," Proc. /E€€ SlCON '97,Apr. 1997,pp. 197-21 1. Department at the University of California, Santa Barbara. At UCSB she is a
[5]S. Murthy and J. J. Garcia-Luna-Aceves, "An Efficient Routing Protocol for member of the Computer Networks and Distributed Systems Laboratory.
Wireless Networks," ACM Mobile Networks and App. 1..Special Issue on Her research interests include routing and multicast for ad hoc mobile net-
Routing in Mobile Communication Networks, Oct. 1996,pp. 183-97. works. She is the recipient of a National Science Foundation Graduate Fel-
[6]A. S. Tanenbaum, Computer Networks, 3rd ed., Ch. 5, Englewood Cliffs, lowship and a University of California Doctoral Scholar Fellowship.
NJ: Prentice Hall, 1996, pp. 357-58.
[7]C. E. Perkins and E. M. Royer, "Ad-hoc On-Demand DistanceVector Routing," CHAI-KEONGTOH [MI ([email protected]) received his Diploma in electronics
Proc. 2nd /€E€ Wksp. Mobile Comp. Sys. and Apps., Feb. 1999,pp. 9C100. and communication engineering with a Certificate of Merit award from the
[8]D. B. Johnson and D. A. Maltz, "Dynamic Source Routing in Ad-Hoc Singapore Polytechnic in 1986,his B.Eng. degree in electronics engineering
Wireless Networks," Mobile Computing, T. lmielinski and H. Korth, Eds., with first class honors from the University of Manchester Institute of Science
Kluwer, 1996, pp. 153-81. and Technology (UMIST) in 1991, and his Ph.D. degree in computer science
[9]J. Broch. D. B. Johnson, and D. A. Maltz, "The Dynamic Source Routing from the Computer Laboratory, University of Cambridge, England in 1996.He
Protocol for Mobile Ad Hoc Networks," IETF Internet draft, draft-ietf- founded and chaired the Mobile Special Interest Group (Mobile SIC) from
manet-dsr-01.txt, Dec. 1998 (work in progress). 1994 to 1996.Before joining Cambridge University, he was a network special-
[lo]V. D. Park and M. S. Corson, "A Highly Adaptive Distributed Routing ist, R&D engineer and technical staff member. At Cambridge, he was an Hon-
Algorithm for Mobile Wireless Networks," Proc. lNFOCOM '97,Apr. 1997. orary Cambridge Commonwealth Trust Scholar and a King's College
[ l l ] M. S. Corson and A. Ephremides, "A Distributed Routing Algorithm for Cambridge Research Scholar. He authored the book Wire/ess ATM and Ad Hoc
Mobile Wireless Networks," ACMIBaltzer Wireless Networks 1.. vol. 1, Networks: Protocols and Architectures (Kluwer, 1996).He serves on the edito-
no. 1, Feb. 1995, pp. 61-81. rial board for ACM Mobile Computing and Communications Review, /E€€ Net-
[12]C-K. Toh, "A Novel Distributed Routing Protocol To Support Ad-Hoc work, and Personal Technologies Journal (Springer Verlag). He is a member of
Mobile Computing," Proc. 7996 /E€€ 75th Annual lnt'l. Phoenix Conf. IEE, USENIX, ACM, Sigma Xi Honor Society, New York Academy of Science,
Comp. and Commun., Mar. 1996,pp. 480-86. American Association for the Advancement of Science (a), and is a Fellow
[13]R. Dube et al., "Signal Stability based Adaptive Routing (SA) for Ad- of the Cambridge Philosophical Society and Cambridge Commonwealth Soci-
Hoc Mobile Networks," /€€€ PerS. Commun., Feb. 1997,pp. 36-45. ety. He is currently an assistant professor with the School of Electrical and
[14]C-K. Toh, "Associativity-Based Routing for Ad-Hoc Mobile Networks," Computer Engineering a t the Georgia Institute of Technology, Atlanta, direct-
Wireless Pers. Commun., vol. 4, no. 2, Mar. 1997, pp. 1-36. ing the Mobile Multimedia and High Speed Networking Laboratory.

IEEE Personal Communications April 1999 55