Methods for Calculating Bandwidth, Delay, and Packet Loss Metrics in

Multi-hop IEEE802.11 Ad Hoc Networks

Liqi Shi1, Abraham Fapojuwo1, Neil Viberg2, Wendy Hoople2, Norbert Chan2
1 2
Department of Electrical and Computer Engineering General Dynamics Canada
The University of Calgary, 2500 University Drive NW 1020 68th Avenue NE
Calgary, AB, Canada T2N 1N4 Calgary, AB, Canada, T2E 8P2
E-mail:{lishi, fapojuwo}@ucalgary.ca {neil.viberg, wendy.hoople, norbert.chan}@gdcanada.com

Abstract — Support of real-time applications in Mobile Ad Hoc methods also serve as useful input for QoS routing of real-time
Networks (MANETs) is very challenging due to the dynamic connectivity in multi-hop MANETs.
characteristics of such networks. This paper studies the problem of The remainder of the paper is organized as follows. Section
voice application support in multi-hop IEEE802.11 ad hoc networks.
We propose reactive-based and non-intrusive methods for calculating
II proposes the bandwidth metric calculation method. In
the quality of service (QoS) metrics of bandwidth, delay and packet Section III, the delay and packet loss measurement schemes
loss that are important for voice support. Results from OPNET based are presented. Section IV gives the numerical results and
simulations show that voice application can be supported in discussion. Section V concludes the paper.
IEEE802.11 MANETs only under light traffic. When the network
traffic is heavy, the calculated delay and packet loss become
II. BANDWIDTH CALCULATION
significantly high due to the hidden node problem, thus making it
difficult to satisfy the voice delay and packet loss objectives in multi- Bandwidth is one of the most limited resources in
hop IEEE802.11 MANETs. MANETs, and most existing QoS routing protocols are based
on this metric. Also, most of the existing protocols assume that
I. INTRODUCTION the network is time slotted and thus corresponding bandwidth
Recently, there is a growing interest in the design of assignment algorithms [9] [10] are proposed. In IEEE802.11
protocols for supporting quality of service (QoS) in mobile ad ad hoc networks, the medium is shared by nodes located in
hoc networks (MANETs). These works mainly focus on the each other’s interference range. Hence, new measurement
network layer, for example, designing routing protocols based schemes are required to obtain the bandwidth information. In
on some QoS metrics [1-3]. However, it is also necessary to the following subsections, two major bandwidth calculation
study the problem of QoS support by considering the schemes – the hello scheme and the listen scheme, are
performance of the underlying medium access control (MAC) reviewed and improved upon for bandwidth calculation in
layer protocol. Since IEEE802.11 [4] is widely deployed in IEEE802.11 ad hoc networks.
wireless equipment for local coverage, this paper investigates
its performance in MANETs that can support real-time A. Hello Scheme
applications. The fundamental access mechanism in the In [11], estimating the available bandwidth via
IEEE802.11 protocol is called distributed coordination neighborhood bandwidth consumption is introduced. That is,
function (DCF). This random access scheme, based on the for any node i in a MANET, it shares the wireless medium
carrier sense multiple access with collision avoidance with all of its neighbors. Thus, the total consumed bandwidth
(CSMA/CA), uses both fixed and random backoff time to in i’s neighborhood, Bi ,consumed , can be written as
transmit/retransmit packets. The IEEE802.11 standard also
defines the point coordination function (PCF) mode, which is Bi ,consumed = ∑ Bj (1)
an access scheme controlled by the access point (AP). In this j∈N ( i )

paper, only the DCF is investigated because the assumed where N(i)={node i and all neighbors of i}, and B is the
scenario is multi-hop MANETs. j
Many papers have studied the performance of the bandwidth consumed by all the existing connections of node
IEEE802.11 protocol in wireless local area networks j, j ∈ N (i ) . Suppose the total bandwidth is Bmax , then the
(WLANs). A Markov chain model of the protocol is given in available bandwidth for node i is
[5]. The effect of contention window size is studied in [6]. Bi , available = Bmax − Bi , consumed (2)
Only few papers attempt to study the IEEE802.11 protocol
In reality, some of i’s neighbors can use the medium
performance in multi-hop MANETs. The effectiveness of the
simultaneously, given that the neighbors and their
RTS/CTS handshake scheme is discussed in [7]. In [8], an
communication peers are outside of each other’s interference
offered load control scheme is used to achieve maximum
range. For example, in Figure 1, nodes A, B, and C are all
throughout for a simple multi-hop scenario where the nodes
neighbors of i, but A-B and C-D can communicate
are in a chain. In this paper, we concentrate on the evaluation
simultaneously since they do not interfere with each other. So,
of bandwidth, delay, and packet loss in multi-hop IEEE802.11
the estimation in (1) is an upper bound for the consumed
ad hoc networks in an attempt to answer the question of
bandwidth and (2) is the lower bound for available bandwidth
whether or not real-time applications can be supported in such
of node i.
networks. In addition, the provided QoS metric calculation

978-1-4244-1645-5/08/$25.00 ©2008 IEEE 103

 requirement at the MAC layer, because the minimum backoff
window size increases exponentially with every
A retransmission.
i The hello scheme is capable of calculating bandwidth in a
C D
scenario where the medium is shared by several nodes.
However, it assumes the consumed bandwidth is fixed, which
B
is not able to reflect the bandwidth dynamics in the real world.
Besides, the accuracy of the calculation also depends on the
Figure 1 A 5-node MANET illustrating the Hello scheme
transmission reliability of hello messages. Hence, the hello
The consumed bandwidth information of each node can be scheme is not suitable for dynamic MANETs.
relayed via HELLO messages. In the above discussions, we
assume that the interference range is the same as the B. Listen Scheme
transmission range, thus one-hop HELLO messages can relay An intuitive way of calculating the bandwidth at the
the information. In case that the interference range is greater IEEE802.11 MAC layer is to “listen” to the medium, and
than the transmission range, multi-hop relay is needed to estimate the available bandwidth during a small period [12].
calculate the available bandwidth. This can be easily achieved in IEEE802.11, since it uses both
To implement the hello scheme in IEEE802.11, protocol physical and virtual carrier sensing (via network allocation
encapsulation overheads and retransmissions must be taken vector, NAV) to determine the availability of the medium. If a
into account. As shown in Figure 2, suppose the length of an node is not in the state of sending or receiving data and the
application data packet is α bits, the protocol overheads above NAV is zero, then the medium is idle, otherwise it is busy.
the IEEE802.11 MAC layer are β bits (i.e., TCP/IP), and the
[A]
MAC and physical layer protocol overhead γ is given by:
[B] [C] [D] [E]

γ=MHDR + CRC + Preamble + PHDR (3)

Listening
then the data packet at the MAC layer becomes α+β+γ bits. interval Call arrival Call release
Let
DIFS BACKOFF DATA SIFS ACK Contention
Tcontrol = TDIFS + TBO + TSIFS + TACK (4)
where Tcontrol is the control overheads in the unit of time to Unit Transmission
Figure 3 Listen scheme
transmit a data packet in IEEE802.11 standard. Assuming the In Figure 3, we show the rules of using the listen scheme to
time to transmit a data packet in Figure 2 is Tdata , then the calculate the available bandwidth. To determine the
bandwidth requirement at the application layer, BApp ,Re q , “listening” period or each measurement cycle, two parameters
are taken into account: the duration of a unit transmission and
becomes BMAC ,Re q at the MAC layer given by: the call arrival/release rate. A unit transmission of IEEE802.11
α + β + γ Tdata + Tcontrol MAC is composed of DIFS, backoff, DATA, SIFS, and ACK,
BMAC ,Re q = BApp ,Re q × × (5) if RTS /CTS (request to send/clear to send) handshake
α Tdata
(optional) is not used. To make the bandwidth estimation
α accurate, the “listening” period should be far greater than the
time needed by a unit transmission, while much smaller than
Above MAC Layer β α the average call interarrival interval or call duration. This is
applicable to applications such as voice. From Figure 3, it is
IEEE802.11 seen that during period [B], bandwidth calculated in [A] is
MHDR β α CRC
used. Since there are no new arriving calls, and no existing
calls are released, the traffic features in periods [A] and [B] are
Preamble PHDR MHDR β α CRC similar, thus we can use the bandwidth calculated in period
[A] during [B]. However, if we use the bandwidth calculated
DIFS BACKOFF (BO) DATA SIFS ACK Contention during period [C] in [D], it is not that accurate since a new call
arrives in [D]. Nevertheless, in terms of efficiency and
Unit Transmission
α: Data packet length
simplicity, the listen scheme is still a good choice for
β: Protocol overheads above the IEEE802.11 MAC layer measuring available bandwidth in IEEE802.11 networks.
MHDR: MAC header
PHDR: Physical layer control protocol header
We propose two methods to implement the listen scheme in
DIFS: Distributed inter-frame spacing IEEE802.11 networks:
SIFS: Short inter-frame spacing
ACK: acknowledgement
z Proactive method: In this method a timer is set to
periodically check the IDLE status of the medium. The
Figure 2 IEEE802.11 control overheads
Finally, assume the number of transmission attempts for medium is idle if no data transmission is in progress. The
available bandwidth proportion can be calculated by n/N,
each data packet is nRT , then the bandwidth requirement for
where n is the number of IDLE samples out of total N
BApp ,Re q at the application layer becomes nRT × BMAC ,Re q at the samples. The only consideration here is the sampling
MAC layer. This value is the minimum bandwidth interval. The interval between two consecutive samples

104
 should be less than half the transmission time of a packet, ( p − p 2 ) / N < mp (9)
such that the samples are not biased (i.e., make sure the
medium busy time is recorded). For example, a 2400-bit Solving Equation (9), the number of samples N is given by:
packet can be transmitted in several milliseconds at a 1− p
N= 2 (10)
data rate of 1 Mbps, and it means that the timer should be m p
set to at least one millisecond or less.
SW router model
z Reactive method: In this approach, medium busy time is Trans delay
t1 t2 t3 t4
either identified by receiver busy status (i.e., received
Processing
power is over a specified threshold) or NAV. One Input queuing delay Output queuing Propagation
delay
important point to note is that the two may be
overlapped. So, the available bandwidth proportion in T t4
Sender Receiver
Try 1
seconds can be calculated by: Try 2
1 − ( NAV + receiver busy time − overlapped time) / T (6)
Both proactive and reactive methods can be used to measure .
DRe trans
the actual available bandwidth effectively, but the reactive . Try n-1
method is more accurate at the cost of higher implementation Try n
complexity than the proactive method. t5
ACK

III. DELAY AND PACKET LOSS CALCULATION Figure 4 Delay measurement in IEEE802.11 networks

Delay and packet loss are two key QoS parameters for real Using eqn. (10), we conclude that this active measurement
time applications. For example, in voice communication, an scheme is not suitable for MANETs. For example, for a link
over-170ms end-to-end delay is said to be harmful to the with 10% mean packet loss rate (p= 10−1 ), we need to send
service quality [13], and a packet loss greater than 5% is
usually unacceptable for voice application [14]. In [15] and 900 samples on that link to get a measurement result whose
[16], one-way delay and round-trip delay measurement standard deviation is within 10% of the mean loss probability
methods are proposed, respectively. However, they are not (i.e., m = 0.1). With each node sending the probes, these can
suitable for IEEE802.11 networks due to significant control cause a huge overhead in MANETs, thus skewing the results.
overheads and complexity. Instead, by analyzing the sources Furthermore, it takes some time to finish the measurements.
of delay at each node in MANETs, we use a software router For example, if one sample packet is sent every second, 15
model to measure the packet node delay. minutes are needed to send 900 samples. The active
The delay at each node in MANETs is composed of input measurement scheme is obviously not suitable for a MANET
queuing delay, processing delay, output queuing delay, with high mobility.
transmission delay, propagation delay, and retransmission To measure the packet loss in MANETs, it is useful and
delay, as shown in Figure 4. To implement the packet delay necessary to understand the factors that cause packet loss.
measurement at a node in IEEE802.11 networks, we need to Unlike the Internet, in which the packet loss is normally
record the time when a packet enters the node (t1), and the caused by congestion, the packet loss in MANETs is due to
time when the data packet is acknowledged (t5) after being many factors. Of all these factors, buffer overflow,
relayed. The node delay for that packet can then be calculated transmission loss, and link breakages are the most dominant.
by t5-t1. Average packet delay in a certain period can also be In addition, a received packet whose delay is over the tolerable
used as a QoS routing metric. delay threshold is also treated as a lost packet. Loss caused by
The packet loss measurement method used in [17] measures over-threshold delay can only be monitored at the receiver,
the packet loss by injecting probing packets into the network. necessitating a feedback message be sent to the source for QoS
The number of probing packets needed to get an accurate purpose. For packet loss caused by link breakages, many
result is analyzed as follows: let schemes have been proposed to switch the connection to
alternative paths before the current path is broken. So, packet
1, packet is lost with probability p loss caused by buffer overflow and maximum retransmissions
Xn =  (7)
0, packet is received with probability 1 − p are the only information that can be obtained from
X N = ( X 1 + X 2 + ... + X N ) / N (8) intermediate nodes. In this paper, we calculate the loss caused
by buffer overflow and maximum retransmissions at
we can see that X N is actually the packet loss rate for N intermediate nodes to evaluate the performance of the
samples. Using the Central Limit Theorem, E( X N )=p, and IEEE802.11 network. Since the active measurement method
needs large amount of probing packets and may skew the
Var( X N )= ( p − p 2 ) / N . Thus, to keep the standard deviation performance of the network, we employ user generated data
of the measurement result within a range relative to the mean, packets to do a reactive measurement instead of injecting
such as mp, where m is the coefficient of variation set to a probing packets into the network.
small value (e.g., m=0.1, 0.01, …). We have:

105
 IV. NUMERICAL RESULTS AND DISCUSSION
Figure 5 shows the simulated scenario implemented in
OPNET, in which SOURCE sends data to DESTINATION via
nodes int_0 and int_1, and SOURCE_0 sends data to
DESTINATION_0 via int_1. The nodes that have no direct
connection line are out of each other’s transmission range.
This is a typical scenario in MANETs. Using this typical
scenario, we evaluate the proposed QoS metric calculation
methods, and determine whether or not voice can be supported
in multi-hop IEEE802.11 ad hoc networks. The configuration
parameters for the simulation are as follows:
z MAC and physical layer: IEEE802.11 - Direct sequence,
1 Mbps, no RTS/CTS, retransmission limits = 7, buffer = Figure 6 Hello scheme and listen scheme: light traffic
256K bits.
z Network layer: Assumed routing protocol is the ad hoc
The results for delay in Figure 7 show that node int_0
on demand distance vector (AODV) protocol.
experiences higher average packet delay (measured every 5
z Application layer: Packet size = 1600 bits, constant bit
seconds) than int_1 although it has lighter traffic. This is due
rate, traffic begins 100 seconds after simulation starting
to higher access delay of int_0, which in turn causes larger
time.
queue size. The higher access delay of int_0 is caused by the
hidden node problem between SOURCE_0 and int_0, which is
common in IEEE802.11 and results in frequent backoffs of the
involved nodes. No packet loss is observed under such a light
traffic.

Figure 5 Simulated scenario

In the first simulated scenario, both SOURCE and

SOURCE_0 transmit data at 16 Kbps, which turns to be 23.48
Kbps at the MAC layer according to Equation (5) and the
control overhead. In Figure 6 the bandwidth measured by the
hello scheme and listen scheme at node int_1 are compared.
According to the hello scheme, the available bandwidth Figure 7 Delay performance of IEEE802.11 networks: light traffic
proportion at int_1 is 0.91 [=(1000-23.48 × 4)/1000] . We use
0.0001 seconds for the proactive method of the listen scheme, In Figure 8, the traffic of SOURCE_0 is increased to 160
which is far more less than the transmission time of one packet. Kbps, thus the available bandwidth portion calculated by hello
The available bandwidth is calculated every five seconds. It is scheme at int_1 is 0.44 [=(1000-23.48 × 4-234.8 × 2)/1000] .
seen that the available bandwidth measured by the two As shown in Figure 8, under high traffic load, the bandwidth
methods of listen scheme matches each other quite well. To consumed by the control messages can be neglected thus the
some extent, the results prove the accuracy of our results of listen scheme and hello scheme are close. However,
measurement methods. As expected, the available bandwidth it is also seen that there are some traffic dynamics reflected
calculated by the hello scheme is fixed because it does not rely gracefully by the listen scheme but not by the hello scheme
on bandwidth dynamics. At a source rate of 16 Kbps, the due to its fixed bandwidth assumption for each connection.
overhead due to hello messages and routing control messages Next, we increase the traffic at both SOURCE and
cannot be neglected. The hello scheme does not take these SOURCE_0 to 160 Kbps. Results in Figure 9 show that node
overheads into account and has an optimized assumption of int_0 experiences much higher delay, queue size, packet loss,
the available bandwidth compared to the listen scheme. and more frequent retransmissions compared to node int_1.
The higher traffic increases the chances of packet collisions
for nodes int_0 and SOURCE_0. Frequent retransmissions at
int_0 cause large backoff periods, very high delay, and packet
loss. An interesting fact is that there is still much bandwidth

106
available while the delay and packet loss are very high at node
int_0. This phenomenon reveals that hidden node problem is
the major factor that degrades the performance of multi-hop
IEEE802.11 ad hoc networks. Increasing transmission rate
(e.g., 54 Mbps) can reduce per packet transmission time, thus
potentially alleviating packet collisions if the same traffic load
is applied. However, this cannot eliminate the hidden node
problem. RTS/CTS handshake is another choice, but it can
only reduce the collisions within the transmitter’s two-hop
range, and cannot deal with any interference from outside of
the range [7]. In addition, RTS/CTS scheme introduces
significant overheads, especially for applications with small
packet size, such as voice (i.e., packet size is 200 bytes or less).

Figure 9 Performance of IEEE802.11 networks: heavy traffic

[4] IEEE Standards Board, "Part 11: Wireless LAN Medium Access Control
(MAC) and Physical Layer (PHY) Specifications," The Institute of
Traffic load (bits/sec) Electrical and Electronics Engineers, Inc., 1997.
210000 [5] G. Bianchi, “Performance analysis of the IEEE 802.11 distributed
205000 coordination function”, IEEE Journal on Selected Area in Comm., Vol .
200000
195000 18, No. 3, March 2000, pp.535-547.
190000
185000
[6] M. Natkaniec and A. R. Pach, “An analysis of the backoff mechanism
180000 used in IEEE 802.11 networks,” in Proc. of fifth IEEE symposium on
175000
170000 Computers and Communications, Antibes-Juan les Pins, France, July 3 -
165000
160000
6 2000, pp. 444–449.
155000 [7] K. Xu, M. Gerla, and B. Sang, “How effective is the IEEE 802.11
RTS/CTS handshake in ad hoc networks,” in Proc. of IEEE
0

5
5

5
0

42

45

47
22

25

27

30

32

35

37

40
10

12

15

17

20

Time (sec) GLOBECOM’02, Vol. 1, November 2002, pp. 72-76.

[8] C. Ng, and S. C. Liew, "Offered Load control in IEEE 802.11 Multi-hop
Figure 8 Hello scheme and listen scheme: heavy traffic Ad-hoc Networks", The 1st IEEE International Conference on Mobile
Ad-hoc and Sensor System, Oct. 2004.
V. CONCLUSIONS [9] C. R. Lin , "On-Demand QoS Routing in Multi-Hop Mobile Networks,"
Proceedings of IEEE INFOCOM 2001, vol. 3, pp. 1735–1744, April
Bandwidth, delay and packet loss measurement methods are 2001.
proposed for evaluating the performance of multi-hop [10] Y. Chen, Y. Tseng, J. Sheu, and P. Kuo, "On-Demand, Link-State,
IEEE802.11 ad hoc networks. Numerical results show the Multipath QoS Routing in a Wireless Mobile Ad Hoc Network,"
efficiency of the metrics in evaluating the protocol’s Proceedings of European Wireless 2002, pp. 135–141, February 2002.
[11] D. H. Cansever, A. M. Michelson, and A. H. Levesque, “Quality of
performance. It is concluded that real time voice connections service support in mobile ad-hoc IP networks”, Proceedings - IEEE
can be supported under a very light traffic load in multi-hop Military Communications Conference MILCOM, v 1, 1999, pp. 30-34.
IEEE802.11 ad hoc networks. However, the delay and packet [12] L. Chen and W. Heinzelman, “QoS-aware Routing Based on Bandwidth
loss are significantly high under heavy traffic load due to the Estimation for Mobile Ad hoc Networks”, IEEE Journal on Selected
Areas in Communications, v 23, no. 3, pp. 561-572, March, 2005.
hidden node problem, this makes it hard to satisfy the [13] Intel in Communications, Whitepaper: “Overcoming Barriers to High-
requirements of real-time voice connections in MANETs. Quality Voice Over IP Deployments,” 2003.
[14] C. Padhye, K. Christensen, and W. Moreno, “A New Adaptive FEC Loss
REFERENCES Control Algorithm for Voice Over IP Applications”, Proceedings of
[1] H. Badis, and K. A. Agha, “Quality of Service for Ad hoc Optimized IEEE International Performance, Computing and Communication
Link State Routing Protocol (QOLSR)”, IETF Internet Drafts, draft- Conference, pp. 307-313, February 2000.
badis-manet-qolsr-01.txt, April 2005. [15] G. Almes, S. Kalidindi, and M. Zekauskas, “A One-way Delay Metric
[2] C. E. Perkins, E. M. Royer, and S. R. Das, "Quality of Service for Ad for IPPM,” IETF RFC2679, September, 1999.
Hoc On-Demand Distance Vector Routing," IETF Internet Drafts, draft- [16] G. Almes, S. Kalidindi, and M. Zekauskas, “A Round-trip Delay Metric
ietf- manet-aodvqos-00. txt, July 2000. for IPPM,” IETF RFC2681, September, 1999.
[3] S. Chen and K. Nahrstedt , "Distributed Quality-of-Service Routing in [17] J. Bolot, “End-to-end packet delay and loss behavior in the Internet”,
Ad Hoc Networks," IEEE Journal on Selected Areas in Communications, Proceedings of ACM SIGCOMM '93, San Francisco, pp.289-298,
vol. 17, no. 8, pp. 1488–1504, August 1999. September 1993.

107