Performance Evaluation of IEEE 802.11 Rate Adaptation Algorithms in Vehicular Networks

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

Performance Evaluation of IEEE 802.

11 Rate
Adaptation Algorithms in Vehicular Networks
Philip Ozorio and Kyeong Soo Kim
System and Process Engineering Centre (SPEC)
College of Engineering
Swansea University
Swansea UK
625750@swansea.ac.uk

AbstractRate adaptation algorithms such as Auto Rate a given link connection to maximize the throughput. In this
Fallback (ARF) and Adaptive ARF (AARF) were designed to work we investigate the performance of two well-know rate
select the optimal bit rate based on the link condition to adaptation algorithms, Auto Rate Fallback(ARF)[6] and
maximize the performance of the wireless network. In vehicular Adaptive Auto Rate Fallback(AARF)[7] in vehicular networks.
networks, the wireless link quality can vary in a short period of For performance evaluation, we implement a simple VANET
time due to packet collision and movement of vehicles at high simulation model using OMNeT++.[8] The simulation looks at
speeds. We used OMNeT++/INET to build a simple the effect of a number of parameters that affects the throughput
infrastructure network model to investigate whether ARF and of the network:
AARF are able to provide higher throughput than no rate
adaptation algorithm (1) in a highly congested network, (2) when Number of vehicles: as the number of wireless nodes
nodes are in motion. We found both ARF and AARF reduced the increases, the number of packet collision will also
number of packet collisions but failed to improve the throughput, increase, reducing the throughput.
which suggests ARF and AARF are not suitable for use in
vehicular networks. Vehicle speed: vehicles do not move at constant speed and
as a result the quality of the link can vary.
Keywords Rate adaptation; ARF; AARF; vehicular networks
Rate adaptation algorithms: these algorithms may improve
I. INTRODUCTION throughput of the network by selecting the optimal bit rate
IEEE 802.11 is a family of medium access control and for transmission.
physical layer specifications for Wireless Local Area Networks A. Software and simulation
(WLANs). The IEEE 802.11 standard provides the basis for This part introduces the two pieces of software used in our
network devices under the Wi-Fi brand. The W-Fi industry
work and the reasons for simulation.
claims to have sold more than one device per person in the
world,[1] and with 802.11g/n being the two most common OMNeT++ stands for Objective Modular Network Testbed
implementations today.[2] The 802.11 standard is constantly in C++.[8] It is a discrete event simulation environment
evolving to provide better performance and capacity to meet primarily used to build communication network simulators.
the ever-increasing demand for connectivity especially with the OMNeT++ alone does not provide any components for
surge of consumer interest in smart devices. Of recent networks but rather providing component architecture for
developments of interest are the Mobile Ad hoc Networks building models. In addition to OMNeT++, we used the
(MANET) and Vehicular Ad hoc Network (VANET).[3] INET[9] framework for OMNeT++. The INET framework is
VANET is a classification of MANET that provides an open-source communication network simulation package
communication between vehicles and fixed equipment such as that has the wireless model and the two rate adaptation
road side units (RSU) over dedicated short range algorithm used to create our VANET model.
communication (DSRC). The objective of VANET is to
There are a few advantages of using a simulation for this
provide road safety measures by allowing the exchange of real-
research over a real network. Firstly, the cost of vehicles and
time data such as vehicular speed and location coordinates
equipment for setting up a vehicular network would cost much
between vehicles. VANET is still at an infancy stage with
more than a computer needed for simulation. Secondly, a real
many networking technologies proposed for use such as
network can become prohibitively expensive if the number of
802.11p and Worldwide Interoperability for Microwave Access
vehicles was to increase and the time needed to setup and run
(WiMAX).[4]
experiments would also increase. Simulation can be used to
802.11g provide multiple physical bit rates from 6Mbps up determine the scalability of the network. Lastly is safety, even
to 54Mbps.[5] A high bit rate can theoretically produce a high in a controlled environment there is an inherent risk of vehicle
throughput when used with high quality links but will produce accident; by using simulator, the user can test any parameters
a lower throughput with lossy or poor quality links. Rate without risk of accidents.
adaptation algorithms try and determine the optimal bit rate for
The disadvantage of simulation is the possibility of A. Mobile Node Model
software bugs which could provide wrong or misleading data The centre of the mobile node is a model of an 802.11
from simulations. However, given the cost and time advantages network interface card. This compound module is configured
of software simulation, software simulation would provide to act as a station with a traffic generator application,
good basis before performing any experimental study in the EtherAppCli connecting to the upperLayerIn of the network
real world. interface card(NIC). LinearMobility module is used model the
B. Rate adaptation algorithms: ARF and AARF mobility of the node and is configured to wrap around to the
opposite border when the node reaches to the end of the
ARF was the first rate adaptation algorithm for 802.11.[7] playground.
ARF was designed to improve the throughput by increasing the
bit rate after a predefined number of successful transmissions B. Road Side Unit Model
in a fixed period of time and switching to a lower bit rate after The road side unit model is simply an 802.11NIC
a few consecutive transmission failures. configured as an access point with the output of the NIC
When the bit rate is increased, a probing packet is sent. connected to a sink module. The road side unit is stationary so
This probing packet must succeed otherwise the rate will be a StationaryMobility module was used here. The objective of
reduced. This poses a problem for links that fade in and out, in this study is to measure throughput, the sink implemented here
particular vehicular networks. Vehicles are constantly moving is a measure of the throughput of all the traffic from the mobile
and the nodes are moving towards and away from different nodes to the road side unit.
roadside units at varying speeds. The mobile node does not C. Playground
know the location of the other nodes nor is unable to predict
them based on previous data. Additionally, ARF and AARF are The network playground is a two dimensional square
receiver based algorithm and is unable to differentiate between measuring 400m by 400m. We model the vehicles and the road
packet collision and transmission failure. As the number of side unit on a flat plane and as the vehicle reaches to the end of
mobile nodes in the network increases, the probability of the border it appears again at the beginning to emulate a
packet collision also increases. constant traffic flow on a motorway.

Even when the link quality remains fairly stable, it is


possible for ARF to produce a lower throughput due to the
algorithms trying to use a higher bit rate than can and causing
many transmission failures. The additional time needed for
retransmit packets reduces the overall throughput. AARF is
designed to address this particular problem by by adapting the
threshold using Binary Exponential Backoff.[7] When a packet
probe fails, the number of successive transmission is multiplied
by two which in effect increases the time until the algorithm
attempts a higher bit rate. This threshold is reset back to the
default value of 10 when the bit rate falls due to two
consecutive transmissions fails.
The effect of AARF is an increase of throughput in stable
conditions over ARF as fewer attempts to transmit at higher bit
rates than the connection allows, consequently fewer failed
transmissions and retransmission.
In the context of vehicular networks, the channel conditions
are not stable as vehicles rarely travel at constant speed and the
Figure 1. The graphical interface of our model. The model consist of a
displacement between cars is continuously changing. By using
variable number of mobile nodes representing by cliHost[*] and a stationary
a simulation we can determine if rate adaptation algorithms can roadside unit represented by ap.)
provide better throughput in the conditions what one might
encounter in a vehicular network. III. SIMULATION SETTINGS
II. METHODOLOGY TABLE I. MODEL PARAMETERS
We present a simple WLAN network built in
MAC Settings
OMNeT++/INET to model our VANET. Our model is based
on the wireless throughput example from the INET framework. rtsThresholdBytes (bytes) 2346
The INET framework provides the implementation of 802.11 cwMinData (number of slots) 31
networking protocols and the rate adaptation algorithm. Our
network was created using OMNeT++ version 4.4.1 and INET cwMaxData (number of slots) 1023
version 2.3 installed on Mac OS X 10.8. cli
reqLength (bytes) 1000
MAC Settings adaptation algorithm on the throughput as the number of
sendInterval (ms) 0.15 vehicles increased.
Rate Adaptation IV. PERFORMANCE EVALUATION
TimerTimeout (number of packets) 15
A. Stationary conditions
minSuccessThreshold (number of packets) 10 25"
maxSuccesThreshold (number of packets) 50 None"
successCoefficient (no units) 2 ARF"

Number"of"Packet"Collision"(x1000)"
20" AARF"
TABLE II. SIMULATION VARIABLES

Variables
**.mac.autobitrate 0,1,2 15"
Number of mobile nodes 1-100

Movespeed (m/s) 5-50 in steps of 5


10"

In the previous sections we briefly mentioned 802.11g/n


being the two most common standard used today. Our VANET
model uses the 802.11g standard. This standard was chosen 5"
over the others because of the availability of hardware and cost.
While this has little effect on our simulation, it would certainly
be a factor if the experiment were to be carried out in the real 0"
world. 1" 4" 16" 64"
Some of the important parameters relevant to the study can Number"of"Vehicles"
be found in Table I. The majority of the internal protocol
Figure 2. Number of packet collisions at ap with stationary vehicles
parameters are the default values defined by the IEEE standard.
In our model, cliEtherApp is set to generate 8000 bit packet at
In Figure 2 we can see a correlation between the number of
an interval of 0.15ms. Equation (1) shows the calculation of the
time period. This should allow the mobile nodes to utilize close packet collisions at ap with the number of vehicles in the
to the 54Mbps available. network. As the number of collision increases, the throughput
decreases. The number of packet collisions was lower when
Packet Length * 1 / Time Period = Output (1) rate adaptation algorithm was enabled.

The parameters in Table II shows the variables that were be 14" None"
used for our experiment. The first experiment was investigating ARF"
the effect of rate adaptation algorithm when there is high
collision. In this case the number of mobile nodes were varied 12" AARF"
Average"throughput"per"car"(Mbps)"

between 1 and 100 then repeated once with each algorithm.


This was conducted with all the mobile nodes when they are 10"
stationary.
The second experiment examines the effect of rate 8"
adaptation algorithm when the mobile nodes are in linear
motion. The number of vehicles was fixed at 20 with an
average velocity ranging from 5 to 50m/s. Vehicles in the real 6"
world rarely travel at a constant velocity so we have applied a
uniform distribution between two intervals to give us an
average velocity. The mobility speed of all the mobile nodes is 4"
independent of each other, which allows them to move towards
or further from each other.
2"
A simulation time limit of 30s is imposed on both
experiments. This allows a good sample size to determine the
throughput of the network. 0"
1" 4" 16" 64"
We used the network model shown in section II with the Number"of"Vehicles"
parameters defined in section III to evaluate the effect of rate
Figure 3. Throughput per vehicle with stationary vehicles
In figure 3 shows the throughput of each vehicle decreases 20"
as the number of mobile nodes increases. In stationary None" ARF" AARF"
conditions we found that while ARF and AARF did help 18"
reduce the number of packet collision but also reduced the

Number"of"Packet"Collision"(x1000)"
throughput of the network substantially. The lower number of 16"
packet collision could be attributed to the lower throughput.
14"
B. Mobile conditions
12"
14" None" ARF" AARF" 10"

8"
12"
6"
Throughput"of"Network"(Mbps)"

10" 4"

8" 2"

0"
6" 5" 10" 15" 20" 25" 30" 35" 40" 45" 50"
Number"of"Vehicles"
4" Figure 5. Number of packet collisions at ap with 20 vehicles in motion

V. CONCLUSION
2" From the results we found that ARF and AARF improved
throughput for a very short duration but as time went on it only
0" reduced the throughput in both stationary and mobile
condition. The throughput when rate adaption is enable was
5" 10" 15" 20" 25" 30" 35" 40" 45" 50"
more than a factor of ten lower than when it is disabled. This
Average"Velocity"(m/s)" indicates that there may be a problem with the simulation or
Figure 4. Thoughput of network with 20 vehicles in motion the implementation of rate algorithm in INET. Another
possibility is that the rate adaptation algorithm is unable to
In figure 4 we can see the throughput fluctuated more distinguish between transmission failures and collision, hence
without rate adaptation but yielded a higher throughput overall failing to adapt correctly for the given link condition.
while the throughput of ARF and AARF were less affected by
the change in velocities but the throughput was more than 10 We also found rate adaptation algorithm increased MAC
times less than without rate adaptation. Figure 5 showed the delay. Concluding from our simulation results, we believe both
number of packet collision was much lower when rate ARF and AARF is not suitable for VANETs because it is
adaptation was enabled. unable to yield long-term performance advantages over no rate
adaptation.
We found the throughput suffered greatly with the rate This work have showed how a VANET can be modeled on
adaption enabled. We investigated further by reducing the OMNeT++ with relative ease and is able to produce theoretical
simulation time limit to a few seconds. We found the results of a large vehicular network at the cost of a personal
throughput with rate adaptation algorithms enabled was higher computer. It should be noted that results from simulation may
during the first two seconds of the simulation but continued to not be a true representation of the real-world. Additional work
decrease as time continued. Our initial prediction was that rate can be carried out to verify the results from our simulation.
adaptation would produce a lower throughput when the link In addition to practical investigation, other rate adaptation
conditions remained stable. However even when conditions algorithm such as SampleRate[10]and other non receiver based
were constantly varied we were not able to produce higher rate adaptation algorithm can be modeled to determine their
throughput than when rate adaptation was disabled. suitability in a VANET.
REFERENCES
[1] Wi-Fi Alliance [Internet].[Las Vegas]:[2014]Wi-Fi connectivity
increases purchase likelihood for smart home devices; Available from:
http://www.wi-fi.org/news-events/newsroom/wi-fi%C2%AE-
connectivity-increases-purchase-likelihood-for-smart-home-devices
[2] C.J. Arges, A survey of IEEE 802.11 protocols. University of Texas,
c2012.
[3] Al-Sultan S, et al. A comprehensive survey on vehicular Ad Hoc [6] A. Kamerman and L. Monteban. WaveLAN-II: A High-performance
network. Journal of Network and Computer Applications, 2013 available wireless LAN for the unlicensed band. Bell Lab Technical Journal pages
from http://dx.doi.org/10.1016/j.jnca.2013.02.036 118-133, Autumn(Fall) 1997
[4] P. Shrivastava, S. Ashai, A.Jaroli and S. Gohil. Vehicle-to-Road-Side- [7] M. Lacage, M.H. Manshaei, T.Turletti, IEEE 802.11 Rate Adaptation:
Unit Communication Using Wimax. International Journal of A Practical Approach Institut National de Recherche en Informatique et
Engineering Research and Applications Vol. 2 Issue4 pages 1653-1655, en Automatique, 2004.
July 2012 [8] A. Varga. OMNeT++: Discrete event simulation system. [Online].
[5] IEEE Std. 802.11-2007, Part 11: Wireless LAN Medium Access Control Available: http://www.omnetpp.org/
(MAC) and Physical Layer (PHY) specifications, IEEE Std. 802.11, [9] Varga et al. INET framework for OMNeT++ 4.0. [Online]. Available:
2007 http://inet.omnetpp.org/
[10] J.C. Bicket, Bit-rate Selection in Wireless Networks, M.S. thesis,
Massachusetts Institute of Technology, 2005.

You might also like