Wireless Sensor Network, 2010, 2, 274-284
doi:10.4236/wsn.2010.24038 Published Online April 2010 (http://www.SciRP.org/journal/wsn)
Comparative Review of QoS-Aware On-Demand Routing
in Ad Hoc Wireless Networks
Ning Zhang, Alagan Anpalagan
WINCORE Lab, Department of Electrical and Computer Engineering, Ryerson University, Toronto, Canada
Received February 23, 2010; revised March 2, 2010; accepted March 5, 2010
Abstract
In this paper, a representative set of QoS models and QoS-aware on-demand routing protocols are reviewed
with emphasis on their ability to support QoS in mobile ad-hoc networks (MANETs) possibly used in WSNs.
In particular IntServ, DiffServ, FQMM, and SWAN QoS models are reviewed followed by different
QoS-aware on-demand routings in MANETs from different perspectives such as the challenges, classifica-
tions, algorithmic aspects in QoS provisions. Tradeoff in providing support to real time (RT) and best effort
(BE) traffic is highlighted. Finally, a detailed and comprehensive comparison table is provided for better un-
derstanding of QoS provision in MANETs.
Keywords: MANET, QoS, QoS Models, QoS-Aware Routing, Real Time Traffic, Best Effort Traffic, Comparison
1. Introduction
Mobile ad-hoc network (MANET) [1] is a collection of
wireless mobile nodes, dynamically forming a temporary
network without pre-existing network infrastructure or
centralized administration. MANETs for wireless sens-
ing and networking have certain unique characteristics
that pose several difficulties in provisioning quality of
service (QoS). They are: dynamic network topology, lack
of precise state information, lack of central control, er-
ror-prone shared radio channels, limited resource avail-
ability and hidden terminal problems [2]. Most routing
protocols for mobile ad-hoc networks, such as OLSR [3],
DSDV [4]; DSR [5] and AODV [6]; and ZRP [7] are
designed without explicitly considering QoS of the
routes they generate. QoS-aware routing requires to find
not only a route from a source to a destination, but a
route that satisfies the end to end QoS requirement such
as bandwidth, delay, jitter or probability of packet loss.
Though QoS in MANETs has been researched exten-
sively, it is a rapidly growing area of research interest
due to the rising popularity and necessity of multimedia
applications.
It is important to understand both QoS models and
QoS routings together in MANETs and there are number
of papers in the literature that discuss comparisons be-
tween them (see [8,9] and references therein). However,
in this paper, they are reviewed with emphasis on differ-
ent perspectives such as the challenges, classifications,
algorithmic aspects for QoS-aware on-demand routing.
The paper is organized as follows: In the next section,
five QoS models are described in detail and then in Sec-
tion 3, challenges in providing QoS are discussed with
the summary of some related work in the literature. Sec-
tion 4 covers various QoS-aware routing algorithms with
their advantages and disadvantages. Finally, the paper
concludes with the summary of the paper.
2. Different QoS Models
A QoS model specifies the network service architecture
that enables us to offer better services than the best effort
(BE) model and plays an important role in providing
QoS support in MANETs. The QoS architecture should
adapt to dynamic topology and time-varying links of
MANETs. In the following, a representative set of QoS
models namely: IntServ, DiffServ, FQMM, and SWAN
are discussed due to their popularity in MANET research
community. The authors in [8,9] provided a good review
of the models and we provide them here for complete-
ness in the following.
2.1. IntServ
The Integrated Service (IntServ) [10] QoS model in-
cludes four components: the classifier, the packet sched-
uler, the admission control routing, and the reservation
setup protocol. An important concept “flow” is defined

N. ZHANG ET AL.275
as distinguishable stream of related datagrams that re-
sults from a single user activity and requires the same
QoS. It is the finest granularity of packet stream distin-
guishable by the IntServ. The basic idea of the IntServ
model is that the flow-specific states are kept in every
IntServ-enabled router. A flow-specific state should in-
clude bandwidth requirement, delay bound, and cost of
the flow. IntServ architecture allows sources to commu-
nicate their QoS requirements to routers and destinations
on the data path by means of a signaling protocol such as
ReSerVation Protocol-RSVP [11]. IntServ proposes two
service classes in addition to BE service. One is guaran-
teed service; the other is controlled load service. The
guaranteed service is provided for applications requiring
strict delay bound. The controlled load service is for ap-
plications requiring reliable and enhanced BE service.
Figure 1 shows how these components work together
to provide integrated services. For each packet, an inter-
net forwarder executes a suite-dependent classifier and
then passes the packet and its class to the appropriate
output driver. A classifier must be both general and effi-
cient. The output driver implements the packet scheduler.
If admission control gives the "OK" for a new request,
the appropriate changes are made to the classifier and
packet scheduler database to implement the desired QoS.
Because every router keeps the flow state information,
the quantitative QoS provided by IntServ is for every
individual flow. In the absence of state aggregation, the
amount of state on each node scales in proportion to the
number of concurrent reservations, which can be poten-
tially large on high-speed links. This model also requires
application support for the RSVP signaling protocol.
IntServ/RSVP model is not suitable for MANETs due
to the resource limitations in MANETs. There are several
factors which prohibit the use of that model over a
MANETs. 1) Scalability: IntServ/RSVP based on per-
flow resource reservation is not appropriate for MANETs
because of the frequently changing topology and limited
resources in MANETs resulting in more signaling over-
head and unaffordable storage and computing process for
mobile nodes. 2) Signaling: The RSVP reservation and
maintenance process is a network consuming procedure.
Thus, RSVP signaling packets will grapple with the data
packets for resources and more specifically for band-
width. This happens because RSVP is an out-of-band
signaling protocol. 3) Router Mechanisms: IntServ im-
poses high requirement on routers. All routers must have
the four basic components: RSVP, admission control
Figure 1. IntServ architecture foundations.
routine, classifier, and packet scheduler. Consequently,
the processing overheads of routers are high which is
undesirable in power-constrained MANETs.
2.2. DiffServ
Differential Service (DiffServ/DS) [12] QoS architecture
is based on a simple model where the traffic entering a
network is classified and possibly conditioned at the
boundaries of the network, and assigned to different be-
havior aggregates. Each behavior aggregate is identified
by a single DS codepoint (DSCP). Within the core of the
network, packets are forwarded according to the per-hop
behavior (PHB) associated with the DS codepoint. The
key components within a differentiated services region
are traffic classification and conditioning functions. Thus,
DiffServ is scalable but it does not guarantee services on
end-to-end basis. This hinders DiffServ deployment in
the Internet as well as in MANETs. Figure 2 shows the
architecture for DiffServ mechanism. The difference
between an edge router and a core router is that the edge
router is required to do the traffic conditioning as defined
by a traffic conditioning agreement between their DS
domain and their peer domain they are connecting to.
DiffServ on the other hand is a lightweight model for
the interior routers since individual state flows are ag-
gregated into a set of flows. This makes routing a lot
more easy in the core of the network.
However, since DiffServ is originally designed for fixed
wired networks, we still face some challenges to imple-
ment DiffServ in MANETs. First, it is ambiguous as to
what the boundary routers in MANETs are. Intuitively, the
source nodes play the role of boundary routers. Other
nodes along the forwarding paths from sources to destina-
tions are interior nodes. But every node should have the
functionality as both boundary router and interior router
because the source nodes can not be predefined. This
drawback would again take us back to the IntServ model
where several separate flow states are maintained, causing
a heavy storage cost in every node. Moreover the concept
of the service level agreement, defined in wire-based QoS
models is not more applicable.
Three QoS classes are defined for the destination to
select the best available path. The first class has the
highest priority and corresponds to applications with
real-time (RT) traffic such as voice. This class is for ap-
plications with high delay constraints. The corresponding
service of this class in DiffServ is referred to as
Classifier Policer Packet Marker Queue Manager
Figure 2. DiffServ architecture foundations.
DS-byte Classifier Queue Manager
Edge Routers
Core Routers
Classifier
Packet
Scheduler
Output Driver
Internet
Forwarder

N. ZHANG ET AL. 276
“expedited forwarding” and in IntServ to guaranteed
service. The second class has less priority than the first
class. It is suitable for applications requiring high
throughput such as video applications. This service class
is referred to as “assured forwarding” in DiffServ and
controlled load in IntServ. The least priority class has no
specific constraint. This class is referred to as best effort
in both architectures. Table 1 shows the defined QoS
classes together with their mappings to IntServ and
DiffServ services. Table 2 is a comparison of IntServ
and DiffServ architecture for QoS.
2.3. FQMM
Flexible QoS Model for MANETs (FQMM) [13] is the
first QoS model proposed for MANETs. It can be viewed
as a hybrid of IntServ and DiffServ model. The basic
idea of this model is that it uses both the per-flow state
property of IntServ and the service differentiation prop-
erty of DiffServ. In other words, this model proposes that
highest priority is assigned per-flow provisioning and
other priority classes are given per-class provisioning.
This model is based on the assumption that not all pack-
ets in the network are actually seeking for highest prior-
ity. FQMM is for small to medium size MANETs using a
flat non-hierarchical topology. This hybrid model defines
three types of nodes, as in DiffServ: a) ingress, if it is
transmitting data, b) core, if it is forwarding data and c)
egress, if it is receiving data. The difference is that in
FQMM the type of a node has nothing to do with its
physical location in the network, since this would not
make any sense in a dynamic network topology.
Figure 3 illustrates the FQMM architecture. A traffic
Table 1. QoS classes and mappings.
Priority Class IntServ DiffServ
1st class e.g. voice,
low delay Guaranteed Expedited For- warding
2nd class e.g. video,
high throughput Controlled Load Assured Forward- ing
3rd class e.g. data, no
constraint BE BE
Table 2. Comparison of IntServ and DiffServ.
Criteria IntServ DiffServ
Granularity Individual flow Aggregate of flows
State in routers Per-flow Per-aggregate
Classification Header fields DS field
Signaling Required(RSVP) Not required
Coordination End-to-end Per-hop
Scalability < # of flows < # of classes
conditioner is put at the ingress node where the traffic
originates. It polices the traffic according to the traffic
profile after a valid route is found. Components of the
conditioner include traffic profile, meter, marker and
dropper. For FQMM, the absolute traffic profile is not
applicable since the effective bandwidth of a wireless
link between nodes is time-varying. Thus, the traffic pro-
file is defined as the relative percentage of the effective
link capacity, in order to keep the differentiation between
classes predictable and consistent under the dynamics of
the network. In addition, the profile should be adaptive to
the dynamics of the network.
FQMM is the first attempt at proposing a QoS model
for MANETs but with the following problems: 1) with-
out an explicit control on the number of services with
per-flow granularity, the scalability problem still exists, 2)
to make a dynamically negotiated traffic profile is a very
difficult problem, 3) it is difficult to code the PHB in the
DS field of IP, if the PHB includes per-flow granularity
considering the DS field is at most 8 bits without exten-
sion. A downside of this approach is that the source sta-
tions have to take great care in regulating their traffic,
since the rate of in-profile traffic must be processable in
all network regions, including bottleneck areas where
traffic from different sources accumulates. However,
FQMM actually lacks the counterpart to DiffServ’s ser-
vice level agreements, and it remains an open question
how the source stations should determine the dynamic
parameter for their token bucket metering.
2.4. SWAN
Service Differentiation Stateless Wireless Ad-hoc Net-
works (SWAN) [14] is a stateless network QoS model
which uses distributed control algorithms with additive
increase multiplicative decrease (AIMD) rate control
mechanism to deliver service differentiation in mobile
wireless ad-hoc networks. The SWAN model includes a
number of mechanisms used to support rate regulation of
BE traffic and admission control regulation of RT traffic,
as illustrated in Figure 4. A classifier and a shaper oper-
ate between the IP and MAC layers. The classifier is
capable of differentiating RT and BE packets, forcing the
shaper to process BE packets but not RT packets. The
Figure 3. FQMM architecture.

N. ZHANG ET AL.277
Figure 4. SWAN model adapted from [14].
shaper represents a simple leaky bucket traffic shaper.
The goal of the shaper is to delay BE packets in confor-
mance with the rate calculated by the rate controller.
What makes such a stateless approach work is that all
nodes independently regulate BE traffic and each source
node uses admission control for RT sessions. When a
new RT session is admitted, the packets associated with
the admitted flow are marked as RT. The classifier looks
at the marking and, if the packet is marked as RT, the
packet will bypass the shaper mechanism, remaining
unregulated. Here, there is an implicit assumption that a
source node regulates its RT sessions based on its admis-
sion control decision.
It is unclear how the amount of bandwidth available
for RT traffic should be chosen in a sensible way.
Choosing larger value results in a poor performance of
RT flows and starvation of BE flows, and choosing it too
low results in the denial of RT flows for which the
available resource would have sufficed. There would
also be no flexibility to tolerate channel dynamics. The
total rate of aggregated RT traffic may be dynamic due
to node changes in traffic patterns and node mobility.
Due to node mobility, for example, intermediate nodes
may need to maintain RT traffic in excess of resources
set-a-side for RT traffic. An intermediate router making
this observation sets the explicit congestion notification
flag in RT packets’ headers. Thus, though SWAN can be
a candidate QoS model, it can not be a complete QoS
solution for a highly dynamic network like MANETs. In
[15], a hybrid approach (incorporating DiffServ) is pro-
posed to achieve performance gain in most of the traffic
load conditions using SWAN model. A comprehensive
simulation study is reported in [16] that is based on the
SWAN model in MANET environment. We can con-
clude that SWAN tries to maintain delay and bandwidth
requirements of RT traffic by admission control of UDP
traffic and rate control of TCP and UDP traffic.
In the above discussed QoS models, certain routing
protocols, algorithms and implementation are not specified,
but the methodology and architecture to provide certain
types of services were presented. There are also other ar-
chitectures that could adopt a hybrid mechanism to guar-
antee the QoS provisions in MANETs. Since achieving
QoS in MANETs not only rely on these models, all the
components such as QoS-aware routing algorithms, QoS
signaling and QoS MAC protocol must also work together
to ensure this. In the next section, different QoS-aware
routing mechanisms are presented and compared.
3. QoS-Aware Routing Mechanism and
RELATED WORK
In any given network, there are two types of flows in
general: BE flows which requires the data to be reliably
delivered to the destination, and QoS flows (such as RT
flows) which apart from reliability, requires some addi-
tional constraints such as available bandwidth, delay, etc.
to be satisfied. Reusing BE routing methods for QoS-
aware routing is not feasible since BE routing performs
these tasks based on a single measure, usually hop-count
while QoS-aware routing, however, must take into ac-
count multiple QoS measures and requirements. This sec-
tion discusses different QoS-aware routings in MANETs
from different perspectives including its challenges,
classifications, algorithms and comparisons.
3.1. Challenges of QoS-Aware Routing
Routing in general consists of two entities, namely the
routing protocol and the routing algorithm. The routing
protocol has the task of capturing the state of the network
and its available network resources, and disseminating
this information throughout the network. The routing
algorithm uses this information to compute shortest paths.
Providing QoS is more difficult for MANETs due to at
least two reasons. First, unlike wired networks, radios
have broadcast nature. Thus, each link’s bandwidth will
be affected by the transmission/receiving activities of its
neighboring links. Second, unlike cellular networks
where only one-hop wireless communication is involved,
MANETs need to guarantee QoS on a multi-hop wireless
path. Further, mobile hosts may join, leave, and rejoin at
any time and at any location; existing links may disap-
pear and new links may be formed “on-the-fly”. All
these raise challenges to QoS-aware routing in MANETs.
Next, we discuss some the challenges in designing rout-
ing protocols.
Dynamic Network Topology: A key challenge in
studying protocol behavior lies in how to represent the
underlying topology and traffic patterns. The constantly
changing and decentralized nature of current networks
results in a poor understanding of these characteristics
and makes it difficult to define a “typical” configuration.
For example, random graphs can result in unrealistically
long paths between certain pairs of nodes, “well-known”
topologies may show effects that are unique to particular
configurations, and regular graphs may hide important
effects of heterogeneity and non-uniformity. The per-
formance of QoS-aware routing depends heavily on the
underlying network topology. The dynamic nature of
Classifier Shaper
Admission
Controller
M
A
C
Admin/reject
Rate
Controller
Mark/unmark Packet delay
ECN Rate
Send/receive
probe Unmarked packet
I
P
Pre-marked/
Unmarked packet
Marked packet

N. ZHANG ET AL. 278
MANETs may make the flow stop receiving QoS provi-
sions due to path disconnections. And also new paths
must be established because of the disconnections and
hence will be causing data loss and delays.
Imprecise State Information: In the link-state rout-
ing algorithms, the source router selects a path based on
the connection traffic parameters and the available re-
sources in the network. The routing protocol distributes
topology and load information throughout the network,
and a signaling protocol for processing and forwarding
connection establishment requests from the source. In
MANETs, the link state changes continuously; hence, the
QoS-aware routing protocols can impose a significant
bandwidth and processing load on the network. Because,
each router must maintain its own view of the available
link resources, distribute link-state information to other
routers, and compute and establish routes for new con-
nections.
Hidden Route Problem: This problem arises at the
time as the route discovery procedure of a QoS-aware
routing protocol is executed. It is because the admission
decision in a route discovery procedure considers only
the local information, e.g., local capacity of the radio
coverage of the node.
Error-Prone Shared Medium: Loss in wired net-
works is typically caused by excessive congestion that
causes packets to be dropped at routers in the network. A
wireless link, however, typically suffers much more loss
due to error-prone shared medium. One cause of loss in
wireless transmission is fading, in which multiple ver-
sions of the same signal are received at the destination. If
these signals are out-of-phase with each other or Dop-
pler-shifted, they can interfere with each other. Other
types of interference may also cause problems in wire-
less transmissions including electrical noise, or possibly
even intentional communication jamming. Propagation
delay can also be a tremendous burden to all communi-
cation, especially to communication that requires a
guarantee on total delay.
Hidden and Exposed Terminal Problem: Consider
the scenario in Figure 5, where there are four common
free time slots between A and B (1, 2, 3, 4) and four free
time slots between B and C (3, 4, 5, 6), if we reserve
slots (1, 2, 3) for A to transmit and slots (4, 5, 6) for B to
transmit, the path bandwidth is only three which is the
maximum number. Suppose there is another pair, D and
E, which are currently using slot 2 to communicate. Then
two cases will occur. If D is a receiver on slot 2, A will
not be allowed to send on slot 2 because otherwise colli-
sion will occur at A. This is the hidden-terminal problem.
So in the example of Figure 5, the common free time
slots between A and B should be reduced to (1, 3, 4) and
the path bandwidth from A to B has to be downgraded to
2 slots. On the contrary, if D is a sender on slot 2, A will
still be allowed to send on slot 2, because this is an ex-
posed-terminal problem. Then the common free time
Figure 5. Hidden and exposed terminal problem in band-
width calculation.
slots between A and B (and thus the path bandwidth)
remain the same. This simple example shows the com-
plication of the bandwidth reservation problem in
QoS-aware routing in MANETs.
Lack of Central Control: Because of the lack of cen-
tral controller which can account for and control
MANETs’ limited resources, nodes must negotiate with
each other to manage the resources required for QoS
routes. This is further complicated by frequent topology
changes. Due to these constraints, QoS-aware routing is
more demanding than BE routing.
Limited Resources Availability: In wireless net-
works, there are additional considerations to be taken
into account. The difficulty of satisfying the QoS re-
quirement is aggravated by further constraints on energy
reserves and available bandwidth, and signal degradation
by noise and limited transceiver resources. Therefore,
instead of a traditional layered network control approach,
a joint optimization scheme affecting both the link and
the routing layer may be necessary.
QoS Signaling Support: INSIGNIA [17] provides
in-band signaling support for QoS in MANETs and it is
more suitable than explicit out-of-band approaches for
supporting end-to-end QoS in highly dynamic environ-
ments where network topology, node connectivity and
end-to-end QoS are highly time-varying. In [18], a hy-
brid QoS model for MANETs, called HQMM, which
combines the per-flow granularity of INSIGNIA [17] and
the per-class granularity of Diffserv [12], is proposed to
provide scalable QoS support for MANETs. In [9], pros
and cons of using cross-layer design approach to QoS
provision in MANETs are discussed.
3.2. Related Work on QoS-Aware Routing
The routing protocols for MANETs may be broadly clas-
sified as table driven protocols [3,4] and on-demand
driven protocols [5,6]. Table driven protocols need to
maintain the global routing information about the net-
work in every mobile node for all the possible
source-destination connection and acquire to exchange
routing information periodically. This kind of protocol
has the property of lower latency and higher overhead.
123456 7 8
C
A
B
D
E
1234567 8
1 2345678
1 2345678
123456 7 8

N. ZHANG ET AL.279
On-demand routing protocol creates routes only when
the source nodes request. When a node requires a route
to a destination, it initiates a route discovery process
within the network. On-demand routing protocols are
characterized as having higher latency and lower over-
head. A majority of existing research on the QoS-aware
routing in MANETs is based on two kinds of route pro-
tocols. However, the table-driven QoS protocols request
globe network state information which is not good for
scalability and on-demand QoS protocols need initiates a
route discovery based on flooding, which are not fit the
dynamic and capability constrain in MANETs. In [19], a
load-balanced AODV (LB-AODV) is proposed to con-
trol the overhead of on-demand routing in MANETs.
QoS-aware routing has received much attention re-
cently for providing QoS in wireless ad-hoc networks
and some work has been carried out to address this criti-
cal issue. Here, we provide a brief review of existing
work addressing the QoS-aware routing issues in wire-
less ad-hoc networks. There have already been several
surveys and overviews regarding the QoS-aware routing
issues and solutions. Authors in [2] summarized the im-
portant QoS-related issues in MANETs in 2001, and the
issues that required further attention. They updated and
expanded their article in 2004 [20]. A fairly comprehen-
sive overview of the QoS in networking could be found
in [21-23]. The main conclusions from the literature are
highlighted below:
1) Many of the underlying algorithmic problems, such
as multi-constraint routing, have been shown to be
NP-complete [20].
2) QoS and BE, routing can only be successfully
achieved if the network is combinatorially stable. The
dynamic topology, the error-prone channel, the lack of
central control and the insecure medium have always
been roadblocks for the development of QoS-aware
routings [22].
3) Different techniques are required for QoS provi-
sioning when the network size becomes very large, since
QoS state updates would take a relatively long time to
propagate to distant nodes [23].
4) The amount of state propagation and topology up-
date information must be kept to a minimum. In particu-
lar, every change in available bandwidth should not re-
sult in updated state propagation [20].
5) QoS-aware routing protocol is designed without
considering the situation when multiple QoS routes are
being setup simultaneously. If two QoS routes cannot be
fully established because they are blocking each other,
both will be deleted. Hence how to setup QoS routes
when there are multiple competing requests needs further
study [24].
6) The protocols should be designed to accommodate
multiple classes of traffic, in particular, to ensure that
lower-class traffic is not starved of network resources in
the presence of RT traffic [23].
4. QoS-Aware On-Demand Routing
Protocols
The problem that concerned the QoS-aware routing pro-
tocol designers was that of discovering the paths that
satisfy the different QoS requirements such as through-
put, delay and jitter in the networks. To find a QoS route
in a MANET is to establish a path that satisfies the QoS
requirement given the knowledge of the available chan-
nel information at each forwarding node. In this section,
some of the main QoS-aware on-demand routing proto-
cols in MANETs are presented and the merits and defi-
ciencies of each protocol are discussed.
4.1. Bandwidth Constraint QoS-Aware Routing
Like DSR [5] and AODV [6], the on-demand QoS-aware
routing protocol [25-27] conforms to a pure on-demand
rule. It neither maintains any routing table nor exchange
routing information periodically. When a source node
wants to communicate with another node for which it has
no routing information, it floods a route request (RREQ)
packet to its neighbors. A bandwidth routing protocol
usually consists of three components: an end-to-end path
bandwidth calculation algorithm to inform the source
node of the available bandwidth to any destination; a
bandwidth reservation algorithm to reserve sufficient
number of free slots for the QoS flow; and a standby
routing algorithm to re-establish the QoS flow in case of
path breaks.
In this protocol, all packets contain following uniform
fields: <packet_type, source_addr, dest_addr, sequence#,
route_list, slot_array_list, data, TTL>. For a source node,
in order to send a stream of packets to a destination node,
a virtual connection (VC), to that node has to be estab-
lished. The VC establishment process includes route
discovery, path bandwidth calculation and bandwidth
reservation components. When a node receives a RREQ
packet in the route discovery process, it records the
status of available slots in the slot_arrary_list. When the
destination node receives one RREQ packet, it returns a
RREP packet by unicasting back to the source following
the route recorded in the route_list. The destination node
selects the path with least cost among them and copies
the fields {route_list, slot_array_list} from the corre-
sponding RREQ packet to the QoS route reply (RREP)
packet and sends the RREP packet to the source along
the path recorded in route-list. As the RREP traverses
back to the source, each node recorded in route_list re-
serves the free slots that have been recorded in the
slot_array_list field. Finally, when the source receives
the RREP, the end-to-end bandwidth reservation process
gets completed successfully and starts sending data
packets in the data phase. The reservations made are soft
state in nature in order to avoid resource lock-up.

N. ZHANG ET AL. 280
The disadvantages of these protocols are: 1) when the
RREP travels back to the source, the reservation opera-
tion may not be successful. This may result from the fact
that the slots which we want to reserve are occupied a
little earlier by another VC or the route breaks. If this is
the case, the route has to be given up and the destination
re-starts the reservation process again along the next fea-
sible route which incurs longer delay; 2) once a VC is
established, the source can begin sending datagrams in
the data phase. At the end of the session, all reserved
slots must be released. These free slots will be contended
by all new connections. However, if the last packet is
lost, we will not know when the reserved slots should be
released; 3) the QoS path discovered in this process may
satisfy the QoS provisions but not necessarily the short-
est path.
4.2. Delay Constraint QoS-Aware Routing
The On-Demand Delay-Constrained Unicast Routing
Protocol (ODRP) is proposed in [28]. For ODRP to work
correctly, each node is required to maintain a distance
vector consisting of |V|–1 entries where |V| is the number
of nodes in the network. The entry for node v at node u
(u! = v) contains the following information: the identifier
of node v, the shortest distance from u to v (in hop count),
and the next hop of u along this path to v. This vector can
be provided by running a proactive wide-area distance
vector routing protocol in the network.
The process of discovering a QoS-aware routing in-
cludes two phases: 1) Probing the feasibility of min-hop
routing. The source sends a packet along the min-hop
routing to the destination and starts a timer. If the
min-hop routing satisfied the delay requirement, this
delay constraint routing has been identified; 2) Destina-
tion initiated route discovery for delay-constraint path.
If the minimum hop path does not satisfy the delay con-
straint, the destination initiates a directed and limited
flood search by broadcasting a RREQ packet. Intermedi-
ate nodes only forward the RREQ with the least delay
value and ignore any further RREQs. When a copy of the
RREQ reaches the source with a path that meets the de-
lay constraint, the route discovery process is complete.
The advantages of this routing protocol are: 1) the path
discovery restricted flooding only when the min-hop
routing does not satisfy the QoS requirements, which
helps to reduce the communication overhead; 2) the
route searching process is restricted and limited in a pre-
determined searching range and each node only forwards
RREQ packet once which further limits the communica-
tion overhead. The disadvantages are: 1) the restricted
searching process may lower the probability of finding a
feasible path; 2) the on-demand nature of route discovery
process leads to higher connection setup time; 3) while
the aim of the directed flooding is to avoid global flood-
ing, thereby reducing overhead compared to protocols
that are based on that, extra overhead is incurred by the
proactive distance-vector protocol which maintains the
routing tables.
4.3. Location Based QoS-Aware Routing
In [29], a predictive location-based QoS-aware routing
protocol is proposed. This protocol includes three com-
ponents: update protocol, predictions (location prediction
and delay prediction) and QoS-aware routing.
The update protocol includes two types of updates.
Type 1 update is generated periodically at a constant
frequency or can vary linearly between a maximum
f(max) and minimum f(min) threshold with the velocity
of the node. Consequently, the distance traveled between
successive Type 1 updates remains constant. Type 2 up-
date is generated when there is a considerable change in
the node’s velocity or direction of motion. In establishing
a connection to a particular destination B, source A has to
first predict the geographic location of the destination B
as well as the intermediate hops, at the instant when the
first packet will reach the respective nodes. Hence, this
step involves a location as well as propagation delay
prediction. The location prediction is used to determine
the geographical location of some node (either an inter-
mediate node or the destination B) at a particular instant
of time t in the future when the packet reaches it.
As a result of the location-resource updates, each node
has information about the entire topology of the network.
It can thus compute a source route from itself to any
other node using the information it has, and can include
this source route in the packet to be routed. The QoS
requirements are in the form of a tuple <estimated dura-
tion of connection, maximum delay, maximum delay jit-
ter>. The maximum delay QoS requirement can be
mapped onto the end-to-end delays observed for the up-
dates from B to A. Thus, given the resource availability
at the nodes and the QoS requirements of the connection,
admission control can be performed. To search for a QoS
path from A to B, A first runs a location-delay prediction
on each node in its proximity list and obtains a list of its
neighbors at the current time. It determines which of
these neighbors have the resources to satisfy the QoS
requirements of the connection. The next step at A’s
network level is to perform a depth-first search for the
destination starting at each of these candidate neighbors
to find all candidate routes. From the resulting candidate
routes, the geographically shortest one is chosen and the
connection is established.
Some of the disadvantages of this protocol include: 1)
it relies on accurate location awareness, which limits its
usefulness to devices that are capable of being equipped
with GPS receivers or such; 2) the update protocol in this
paper involves flooding of location and resource infor-

N. ZHANG ET AL.281
mation pertaining to a node to all the other nodes in the
network. Ordinarily, such a full flooding of the network
involves a very large overhead. However, with schemes
such as the multipoint relay scheme, the overhead asso-
ciated with flooding can be considerably reduced.
4.4. Hierarchical Routing: CEDAR
CEDAR [30], a core-extraction distributed ad-hoc rout-
ing algorithm for QoS-aware routing in ad-hoc network
environments, has three key components: 1) the estab-
lishment and maintenance of a self organizing routing
infrastructure called the core for performing route com-
putations; 2) the propagation of the link state of high
bandwidth and stable links in the core through in-
crease/decrease waves; and 3) a QoS-route computation
algorithm that is executed at the core nodes using only
locally available state.
Core Extraction: The core structure is used to limit
the number of nodes that must participate in the ex-
change of topology and available bandwidth information.
The goal of setting up the core is to proactively create a
core set such that every node is either a core node or a
neighbor of a core node. As the route computation is
done by the core nodes, minimizing the number of core
nodes is desirable. Since core computation is local, it
makes core computation in CEDAR scalable as the core
can be computed in a constant amount of time. When a
node is electing a dominator, it gives preference to core
nodes already present in its neighborhood (including
itself). This provides stability to the core computation
algorithm, though it might have implications on the op-
timality of the number of core nodes. Each core node
maintains local topology information and performs route
discovery, route maintenance and call admission on be-
half of these nodes.
Core Broadcast: In order to achieve efficient core
broadcast, each node temporarily caches every request to
send (RTS) and clear to send (CTS) packet that it hears
on the channel for core broadcast packets only. The pur-
pose of caching RTS/CTS is to use them for the elimina-
tion of duplicate packet reception for broadcasts.
QoS State Propagation: To propagate state informa-
tion (available bandwidth) among the core nodes, in-
crease waves and decrease waves are used. These waves
are generated when a core node’s available bandwidth
has changed by a threshold value. A slow-moving in-
crease wave denotes an increase of bandwidth on a link,
and a fast-moving decrease wave denotes a decrease of
bandwidth on a link. For low-bandwidth links, it makes
sense to have as few nodes as possible contending for the
link, while for stable high bandwidth links, it makes
sense to have as many core nodes as possible know about
the link in order to compute good routes. In other words,
the maximum distance that the link state can travel (i.e.
the time-to-live field) is an increasing function of the
available bandwidth of the link. And because every core
node that caches information corresponding to a link can
potentially use the bandwidth of the link, the number of
core nodes that cache the state of a low bandwidth link
should be less compared to a stable high bandwidth link
to reduce the contention for a low bandwidth link.
QoS-Aware Routing Setup: Briefly, QoS route com-
putation in CEDAR is an on-demand routing algorithm
which proceeds as follows: when a source node s seeks
to establish a connection to a destination node d, pro-
vides its dominator node dom(s) with a (s, d, b) tuple,
where b is the required bandwidth for the connection. If
dom(s) can compute an admissible available route to
using its local state, it responds to immediately. Other-
wise, if dom(s) already has the dominator of d cached
and has a core path established to dom(d), it proceeds
with the QoS route establishment phase. If dom(s) does
not know the location of d, it first discovers dom(d), si-
multaneously establishes a core path to d , and then initi-
ates the route computation phase. A core path from s to d
results in a path in the core graph from dom(s) to dom(d);
dom(s) then tries to find the shortest-widest-furthest ad-
missible path along the core path. Based on its local in-
formation, dom(s) picks up the farthest reachable domain
until that which it knows is an admissible path. It then
computes the shortest-widest path to that domain, once
again based on local information.
The advantages of this routing protocol includes: route
computation does not involve the maintenance of global
state and only a few nodes are involved in state propaga-
tion and route computation. If the topology stabilizes,
then routes will converge to the optimal routes. Disad-
vantages include: As far as the nature of state maintained
at each core node is concerned, at one extreme is the
minimalist approach of only storing local topology in-
formation at each core node. This approach may result in
a poor routing algorithm (i.e., the routing algorithm may
fail to compute an admissible route even if such routes
exist in the ad-hoc network). At the other extreme is the
maxima list approach of storing the entire link state of
the ad-hoc network at each core node. This approach
computes optimal routes for stable networks, but incurs a
high state management overhead for dynamic networks
and potentially computes stale routes based on an
out-of-date cached state when the network dynamics are
high.
4.5. Application-Aware QoS-Aware Routing
A unique approach to QoS-aware routing is presented in
[31]. It is unique because instead of using lower layer
information, it is based on the aid of the transport layer.
The protocol assumes the use of RT transport protocol
(RTP) and the RT streams are delivered in the RTP

N. ZHANG ET AL. 282
packets. The delay between two nodes is estimated sta-
tistically by examining the difference between time-
stamps on transmission and receipt of RTP packets be-
tween those two nodes. The delay variance is also calcu-
lated. Furthermore, each node records the throughput
requirement of RTP sessions which are flowing through
it. Subtracting the total of these throughput values from
the raw channel capacity gives an estimate for the total
remaining capacity at that node.
Figure 6 shows a MANET including eight mobile nodes.
The dashed line between two nodes represents the wireless
connection. The number tag of each dashed line denotes the
estimated transmission delay in the unit of 10 ms.
The route discovery is performed in the following
steps:
Step 1: Using Floyd's algorithm, define a set I=
{} including all the routes with the short- est
delays satisfying the delay requirements.
k
RRR ,...,, 21
Step 2: Select the subset A whose elements satisfy the
bandwidth requirement. If set A is null, then go to Step
4.
Step 3: From set A, select the route R with the mini-
mum variance of the transmission delays during a prede-
fined period.
Step 4: Select the route R with the maximum allocated
available bandwidth. If there is sufficient available
bandwidth for a multimedia application, the most robust
QoS route is selected using this scheme. If there are no
routes that meet the bandwidth requirement, the route
with the highest available channel capacity, which satis-
fies the delay constraint, is selected.
Using Floyd’s algorithm, we could compute four
routes that satisfy the delay requirement from Vs to Vt: 1,
Vs→V2→Vt; 2, Vs→V1→V2→Vt; 3, Vs→V2→V4→ Vt
and 4, Vs→V3→V2→Vt. From Step 2, we could elimi-
nate those routes that do not satisfy the bandwidth re-
quirement. We assume that route 1 and 2 can satisfy the
bandwidth requirement. Then from Step 3, we could
choose the route that has the minimum delay variance as
the QoS route. If none of the routes satisfy the bandwidth
requirement, the route with the maximum available
bandwidth will be selected.
A major advantage of this routing protocol is that no
extra overhead is incurred for QoS-aware routing, since
the existing transport layer packets are used for QoS
2
1.5
5.5
1.5
1 1
V1
V6
V4
V5
4
6
2.2
V2
V3
2 1
2
V
t
Vs
Figure 6. Network topology of application-aware routing
adapted from [31].
Table 3. Comparison of QoS-aware on-demand routing protocols.
Routing Algorithms QoS Metrics Architecture
& Reactive Network/Node InformationMAC Layer Other Assumptions
Bandwidth guaran-
teed Routing (CCBR)
[23]
Bandwidth Flat/Proactive Time slot schedule
Neighbor nodes status
CDMA over
TDMA; re-
source
reservation
DSDV routing, call
admission control
On-demand
QoS-aware routing
[32]
Bandwidth Flat/Reactive
Node states
Neighbor nodes status
CDMA over
TDMA; re-
source
reservation
AODV routing
Delay constrained
routing (ODRP)
[25]
Bounded
delay Flat/Reactive
Distance vector consisting of
|V|-1 entries (identifier of V,
shortest path, next hop)
Resource
reservation
AODV routing but pro-
active state dissemina-
tion
Bandwidth Reserva-
tion Multi-path
QoS-aware routing
[27]
Bandwidth Flat/Reactive Link state information Resource
reservation AODV routing
Predictive loca-
tion-based
QoS-aware routing
[29]
Improved
link and path
longevity
Flat/Reactive Node relative positions
and velocities None
Relative location aware-
ness; relative speed
awareness;
source-routing
CEDAR (Core Ex-
traction Distributed
Ad-hoc Routing)
[26]
Bandwidth Hierarchical/
Partially Link residual capacity
Link residual
capacity
estimation
RTS/CTS is cached for
the purpose of core
broadcasting
Application-aware
QoS-aware routing
[31]
Bounded
delay Flat/Reactive RTCP information None RTP is needed

N. ZHANG ET AL.283
metric estimation. Additionally, both delay and through-
put constraints may be considered. However, the use of
RTP is assumed, and therefore the range of application
scenarios for this protocol is obviously limited.
Comparison of QoS-aware Routing Protocols
There are different ways to classify the QoS-aware
routing protocols in MANETs. Some classify the proto-
cols by the network topology (flat, hierarchical, hybrid).
Some classify the protocols by different approaches to
solve the QoS issues (ticket-based probing, predictive,
more node state information). Some classify the proto-
cols by route discovery approach (proactive, reactive,
hybrid). Other typical classifications include by the in-
teraction with MAC layer (independent or dependent),
and also by the QoS requirements (delay, bandwidth,
security, energy). In this paper, the classification of
QoS-aware routing protocols is based on the approaches
to QoS-aware routing in MANETs. Table 3 lists the rep-
resentative QoS-aware routing mechanisms discussed in
this paper. It includes the QoS metrics, the node infor-
mation, the requirement from MAC layer and other as-
sumptions to make the protocols feasible.
5. Conclusions
In this paper, we presented a representative set of
QoS-models and QoS-routings for MANETs with an
emphasis on QoS-aware on-demand routing and their
support for QoS provision. Although most of the re-
search focus on different problems, they are related to
each other and have to deal with some common difficul-
ties, which include mobility, limited bandwidth and
power consumption, and broadcast characteristic of radio
transmission in MANETs. A detailed and comprehensive
comparison table is also provided for better understand-
ing of QoS provision in MANETs through on-demand
routing mechanisms. Cross-layer approach to QoS provi-
sion in MANETs has to be carefully researched to ex-
ploit the layer interactions effectively. Though layers
with strict and well defined boundaries lend themselves
to modular design/testing and interoperability, they could
pose some challenges in the overall system implementa-
tion [33,34].
6. References
[1] S. Corson and J. Macker, “Mobile Ad-hoc Networking
(MANET): Routing Protocol Performance Issues and
Evaluation Considerations,” Network Working Group
Request for Comments, No. 2501, 1999.
[2] S. Chakrabarti and A. Mishra, “QoS Issues in Ad-Hoc
Wireless Networks,” IEEE Communications Magazine,
Vol. 39, No. 2, February 2001, pp. 142-148.
[3] P. Jacquet, et al., “Optimized Link State Routing Protocol
for Ad-Hoc Networks,” IEEE INMIC Multi Topic Con-
ference, Lahore, December 2001, pp. 62-68.
[4] C. E. Perkins and P. Bhagwat, “Highly Dynamic Destina-
tion-Sequenced Distance-Vector Routing (DSDV) for
Mobile Computers,” ACM press, Vol. 24, No. 4, October
1994, pp. 234-244.
[5] J. Broch, et al., “The Dynamic Source Routing Protocol
for Mobile Ad-hoc Networks,” IETF Internet-Draft,
March 1998.
[6] C. E. Perkins, “Ad-hoc On-demand Distance Vector Rou-
ting,” IETF Internet-Draft, November 1997.
[7] Z. J. Haas and M. R. Pearlman, “ZRP: A Hybrid Frame-
work for Routing in Ad-Hoc Networks,” Ad-hoc Net-
working, 2001, pp. 221-253.
[8] K. Wu and J. Harms, “QoS Support in Mobile Ad Hoc
Networks,” Interdisciplinary Journal of Crossing Bounda-
ries, Vol. 1, No. 1, 2001, pp. 92-106.
[9] N. Sarma and S. Nandi, “Enhancing QoS Support in Mo-
bile Ad Hoc Networks, Advances in Computer, Informa-
tion, and System Sciences and Engineering,” Springer,
2006, pp. 267-273.
[10] R. Braden, D. Clark and S. Shenker, “Integrated Services
in the Internet Architecture: An Overview,” Network
Working Group Request for Comments, June 1994.
[11] L. Zhang, et al., “RSVP: A New Resource ReSerVation
Protocol,” IEEE Network, Vol. 7, No. 5, September 1993,
pp. 8-18.
[12] S. Blake, “An Architecture for Differentiated Services,”
Network Working Group Request for Comments, De-
cember 1998.
[13] X. Hannan, et al., “A Flexible Quality of Service Model for
Mobile Ad-Hoc Networks,” IEEE Vehicular Technology
Conference, Vol. 1, No. 15-18, May 2000, pp. 445-449.
[14] G. Ahn, et al., “Supporting Service Differentiation for RT
and Best effort Traffic in Stateless Wireless Ad-Hoc Net-
works (SWAN),” IEEE Transaction on Mobile Comput-
ing, Vol. 1, No. 3, July-September 2002, pp. 192-207.
[15] N. Sarma and S. Nandi, “QoS Support in Mobile Ad Hoc
Networks,” IFIP International Conference on Wireless and
Optical Communications Networks, April 2006, pp. 1-5.
[16] N. Zhang and A. Anpalagan, “Sensitivity of SWAN QoS
Model in MANETs with Proactive and Reactive Routing:
A Simulation Study,” Telecommunication Systems, 2008.
[17] S-B. Lee and A. T. Campbell, “INSIGNIA: In-band Sig-
naling Support for QoS in Mobile Ad-Hoc Networks,”
International Workshop on Mobile Multimedia Commu-
nication, 1998.
[18] Y. He and H. Abdel-Wahab, “HQMM: A Hybrid QoS
Model for Mobile Ad-Hoc Networks,” IEEE Symposium
on Computers and Communications, June 2006, pp. 194-
200.
[19] J. H. Song, V. W. S. Wong, V. C. M. Leung, “Efficient
On-demand Routing for Mobile Ad-Hoc Wireless Access
Networks,” IEEE Journal on Selected Areas in Communi-
cations, Vol. 22, No. 7, September 2004, pp. 1374-1383.

N. ZHANG ET AL. 284
[20] S. Chakrabarti and A. Mishra, “Quality of Service Chal-
lenges for Wireless Mobile Ad-Hoc Networks,” Wiley J.
Wireless Communications and Mobile Computing, Vol. 4,
March 2004, pp. 129-153.
[21] Z. Chenxi and M. S. Corson, “QoS-Aware Routing for Mo-
bile Ad-Hoc Networks,” IEEE International Conference on
Computer Communications, Vol. 2, pp. 958-967.
[22] R. Renesse, et al., “QoS Conflict Resolution in Ad-Hoc
Networks,” IEEE International Conference on Commu-
nications, Vol. 8, June 2006, pp. 3826-3831.
[23] T. B. Reddy, et al., “Quality of Service Provisioning in
Ad-Hoc Wireless Networks: A Survey of Issues and So-
lutions,” Ad-Hoc Networks, Vol. 4, No. 1, January 2006,
pp. 83-124.
[24] C. R. Lin, “QoS-aware Routing in Ad-Hoc Wireless Net-
works,” 23rd Annual Conference on Local Computer
Networks, October 1998, pp. 31-40.
[25] C. R. Lin and J. Liu, “QoS-Aware Routing in Ad-Hoc
Wireless Networks,” IEEE Journal on Selected Areas in
Communications, Vol. 17, No. 8, August 1999, pp. 1426-
1438.
[26] C. R. Lin, “On-demand QoS-aware Routing in Multihop
Mobile Networks,” IEEE International Conference on
Computer Communication, Vol. 3, April 2001, pp. 1735-
1744.
[27] W.-H. Liao, Y.-C. Tseng, S.-L. Wang and J.-P. Sheu, “A
Multi-path QoS-Aware Routing Protocol in a Wireless Mo-
bile Ad-Hoc Network,” 1st International Conference on
Networking-Part 2, London, Vol. 2094, 2001, pp. 158-167.
[28] B. Zhang and H. T. Mouftah, “QoS-Aware routing for
Wireless Ad-Hoc Networks: Problems, Algorithms and
Protocols,” IEEE Communications Magazine, Vol. 43,
No. 10, October 2005, pp. 110-117.
[29] S. H. Shah and K. Nahrstedt, “Predictive Location-Based
QoS-Aware routing in Mobile Ad-Hoc Networks,” IEEE
International Conference on Communications, New York,
Vol. 2, May 2002, pp. 1022-1027.
[30] P. Sinha, R. Sivakumar and V. Bharghavan, “CEDAR: A
Core-Extraction Distributed Ad-Hoc Routing Algorithm,”
IEEE Journal of Selected Areas of Communications, Vol.
17, No. 8, August 1999, pp. 1454-1465.
[31] W. Min and K. Geng-Sheng, “An Application-Aware
QoS-Aware routing Scheme with Improved Stability for
Multimedia Applications in Mobile Ad-Hoc Networks,”
IEEE Vehicular Technology Conference, Stockholm, Vol.
3, No. 25-28, September 2005, pp. 1901-1905.
[32] S. Chen and K. Nahrstedt, “Distributed Quality-of-service
Routing In Ad-Hoc Networks,” IEEE Journal of Selected
Areas of Communications, Vol. 17, No. 8, August 1999,
pp. 1488-1505.
[33] H. Tian, et al., “CLA-QOS: A Cross-Layer QoS Provi-
sioning Approach for Mobile Ad-Hoc Networks,” IEEE
TENCON, November 2005, pp. 1-6.
[34] Q. Liu, et al., “A Cross-Layer Scheduling Algorithm with
QoS Support in Wireless Networks,” IEEE Transactions
on Vehicular Technology, Vol. 55, No. 3, May 2006, pp.
839-847.