See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/221000955

Improving performance of MAC layer by using congestion control/avoidance

methods in wireless network

Conference Paper · January 2001

DOI: 10.1145/372202.372390 · Source: DBLP

CITATIONS READS
13 307

3 authors, including:

Hamid Sharif Guevara Noubir

University of Nebraska at Lincoln Northeastern University
313 PUBLICATIONS 5,409 CITATIONS 155 PUBLICATIONS 3,010 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Guevara Noubir on 28 May 2014.

The user has requested enhancement of the downloaded file.

Improving Performance of MAC Layer by Using Congestion
Control/Avoidance Methods in Wireless Network*

Song Ci Hamid Sharif Guevara Noubir

Universityof Nebraska-Lincoln Universityof Nebraska-Lincoln CSEM
EE Deptartment CEEN Department Network Department
Lincoln, NE68588 Omaha, NE68182 Neuchatel, Switzerland
[email protected] hsharif@ unomaha.edu [email protected]

Keywords its wired counterpart. In addition, wireless channel is error-

Wireless LAN, QoS, MAC, IEEE802.11, Adaptive algorithms prone, bursty channel, thus this makes the design of W L A N
M A C layer protocol more difficult. Moreover, hidden sta-
tion problem increases the design complexity [2]. Similarly,
ABSTRACT in connectionless wired packet network, the end user can
In this paper, adaptive fragment algorithms for IEEE 802.11
not detect whether there is congestion happened inside the
wireless LAN are proposed a n d studied. This work is in-
network due to finitebuffers and unpredictable load. There-
spired by studying end-to-end congestion control/avoidance
fore, the design of both categories algorithms has to face the
m e t h o d s used in t r a n s p o r t layer. The throughput perfor-
problem which is how to allocate the network resource dy-
mance of proposed algorithms for IEEE 802.11 wireless LAN namically.
is simulated under different channel quality scenarios. Ac-
cording to the simulation results, the adaptive algorithms The M A C layer protocol in W L A N will try to supply a col-
designed by using the m e t h o d s behind end-to-end conges-
lision flee access method for multiple users and it also has
tion control/avoidance algorithms can improve the channel
to use some means to improve link reliability to avoid re-
t h r o u g h p u t a n d reduce the end-to-end delay, even the chan-
transmissions happen because re-transmissions will be very
nel is very noisy. expensive in W L A N . In order to satisfy these new require-
ments for M A C layer protocol in W L A N , I E E E 802 project
1. INTRODUCTION proposes the I E E E 802.11 W L A N standard. In this stan-
In recent years, there is an increasing need for more wire- dard, Distributed Coordination Function (DCF) and Point
less d a t a services, such as voice, email, video and audio, ac- Coordination Function (PCF) are specified to supply a col-
cessed by laptop, paimtop a n d so on. As one of "last-hop" lision free multiple access environment. A n d besides, M A C
solutions, wireless LAN ( W L A N ) is adopted to connect mo- layer A C K (Acknowledgement) and N A V (Network Alloca-
bile devices to high speed wired networks. In studying the tion Vector), new back-off method, R T S (Request to Send),
issues faced by MAC algorithms, in some senses, they are C T S (Clear to Send), M A C layer fragmentation and de-
very similar to issues faced by end-to-end congestion control fragmentation are specified to increase throughput and re-
algorithms in connectionless wired packet networks. Before liability of communications. The goal of adopting M A C
this is explained further, it is necessary to compare and con- layer fragmentation and de-fragmentation is to avoid re-
t r a s t problems faced by these two categories of algorithms. transmissions from higher layers. For example, due to delay
and window size, the cost of T C P layer re-transmission will
First, the goal of these two categories of algorithms is to be much higher than that of M A C layer re-transmission.
improve network throughput. In W L A N , the carrier sens- Similarly, the cost of congestion in connectionless packet
ing m e t h o d is much different from the way used in wired network is also very high, thus, in transport layer, end-
shared-media network like Ethernet. This is because colli- to-end congestion control algorithms have to be adopted.
sion is detected right away in Ethernet, but this is impossible There are some other congestion control/avoidence means
in W L A N even though it is also a shared-media network like like fair queuing and hop-by-hop flow control, but in this
paper, only end-to-end congestion control/avoidence algo-
*This paper is partially sponsored by M o C o R e P r o Project, rithms adopted by transport layer will be discussed.
University of Nebraska-Lincoln.
The rest of this paper is organized as follows. In the next sec-
Permission to make digitalor hard copies of all or part of thiswork for tions, we discuss the possible application of T C P congestion
personal or classroom use is granted without fee provided that copies are
not made or distributed for profitor commercial advantage and that copies control/avoidance algorithms in IEEE 802.11 W L A N . In
bear this notice and the full citation on the first page. To copy otherwise,to section 3, four adaptive fragmentation algorithms for I E E E
republish, to post on sm-versor to redistribute to lists, requires prior specific 802.11 W L A N are proposed in heuristics of studying T C P
permission and/or a fee. congestion control/avoidance algorithms. In section 4, the
SAC 2001, LainVegas, NV simulation model is described. The simulation results and
© 2001 ACM 1-58113-287-5/01/02...$5.00 analysis are presented in section 5. Finally, we will conclude
with a summary.

420
2. PROBLEM DESCRIPTION TCP layer, packet loss is thought as the result of congestion,
In IEEE 802.!1 WLAN, there is a new enhancement for QoS similarly, packet loss is thought as bad packet received at re-
provision, i.e., MAC layer fragmentation and de-fragmentation. ceiver or packet destroyed by hidden stations in WLAN. In
The advantage of using fragmentation and de-fragmentation TCP end-to-end congestion control/avoidance algorithms,
is that when end users send long data packets, the frame er- window size is used as a unit of controlling, i.e., through
ror rate is very large because of the characteristics of wireless controlling the window size in sender or receiver or both,
channel. Thus, this causes a large number of re-transmission congestion will be controlled or avoided.
and then generate much overhead. According to [6, 7], the
mean attempts of transmission(?,) is In WLAN, packet size or fragment size or both are controlled
1 in a similar ways to reduce the frame error rate. Here, four
~= (1-py) (1) adaptive fragmentation algorithms are proposed in heuris-
tic of studying congestion control/avoidance algorithms and
and the network throughput(C) is their performance will be compared with the fragment adap-
tive algorithm using fragment size back-off [4]. The algo-
1 rithm 1 described below is designed by using the method be-
¢= r x k xps (2) hind slow-start congestion control algorithm widely adopted
Where, in transport layer [9].
r : the coding rate such as 1, ~,
~ 5 etc.
k : is equal to log 2 M, and M means M-array modulation 3.1 Adaptive Fragmentation Algorithm 1
scheme.
p / : error probability of frame.
If(ACK lost or ACK time out)
In this work, we choose no coding and BPSK modulation Ok+l = Ok + 2;
case, i.e., r is set to 1 and k is set to 1. As P f = 1 - (1 --pb) N , if (0k+l < e)
Pb is bit error probability and N is the length of frame. Ok+l = e;
Therefore, Else
Ok+~ = Ok X2;
1
7 -- (1 - - p b ) N (3) if (0k+l > 6)
Ok+l = 6;
In this work, another type of throughput equation is used
for simplification purpose,
In the above algorithm, the e is the optimal fixed fragment
¢ = V (4) size in corresponding to different channel quality. The O is
T
the adaptive fragmentation threshold used by sender.e is the
Here, G is the total number of bytes received successfully total size of the long frame sent using fragmentation algo-
and r is total simulation time. Therefore, using fragmen- rithms. If ACK is lost or time out, the adaptive fragmen-
tation will lower the frame error rate and then increase the tation threshold will be back-off by half, i.e., in time k + 1,
network throughput. the threshold will be half of the threshold in time k. On the
other hand, if ACK is received, the fragmentation threshold
The other advantage of using MAC layer fragmentation is will be doubled in time k + 1. The fragraentation threshold
to reduce the total time of transmitting a longer packet. will not be increased more than the frame size and it will
In this case, the sender is just waiting for a Short Inter- also not be decreased less than the optimal fragmentation
Frame Space (SIFS) after receiving an ACKnowledgement size.
(ACK) of fragment from receiver rather than a Distributed
Interframe Space (DIFS)(according to the specification of
IEEE 802.11 W L A N standard, DIFS is always larger than 3.2 Adaptive Fragmentation Algorithm 2
SIFS). As wireless link is error prone and time-varying, this
fragment size should not be a constant value like in wired If(ACK lost or ACK time out)
network. We believe the adaptive algorithm enhances the generate n E [1, v]with uniform dist.;
throughput and reduces the end-to-end delay. Ok+l = Ok + n;
if (0k+l < e)
Even though the IEEE 802.11 specifies the fragmentation Ok+l = e;
approach, no adaptive fragmentation algorithm is given and Else
there is no research work in this topic so far. As it is an- generate m 6 [1, w] with uniform dist.
alyzed in Section 1, we suggest that the design methods Ok+l = O k x m;
behind T C P end-to-end congestion control/avoidance algo- if (Ok+l > e)
rithms could help to improve the performance the MAC 0k+l = 6;
layer protocol of WLAN.

3. ADAPTIVE ALGORITHMS SUMMARY In the above algorithm, fragment size is increased or de-
The similarity between issues faced by T C P and issues faced creased exponentially in a random way. This algorithm is in-
by WLAN is that the end user sends a packet into network spired by the random back-off algorithm employed by some
without reservation and then wait an event happens. In MAC layer protocols. The discussion and comparison of

421
these two algorithms appear in [4]. v and to axe the max- aOl / uOl
imum back-off window size and n and m axe the back-off
step used in time k + 1.
all

3.3 Adaptive Fragmentation Algorithm 3

al0/ul0
If(ACK lost or ACK time out)
generate n E [1, v]with uniform dist.; Figure 1: C h a n n e l state diagram for Gilbert-Elliot
Ok+l = Ok + n; Bursty Channel Model
if ( 0 k + 1 < e)
O k + l = e;
Else
Ok+~ = Ok + 6; 4. SIMULATION MODELING
if (0k+l > ~) In this paper, the wireless channel model is characterized
by Gilbert-Elliot Model [10]. This model uses a two-state
ergodic Markov chain whose steady states here are AWGN
channel with different N~--~. We choose 30dB difference be-
In above, fragment size is decreased in a random exponential tween two steady states o~ Gilert-Elliot burst channel model,
way but increased in a additive way. Here 6 is the fragment which are 10 -5 in good state and 10 -2 in bad state. In
increase step. Figure.I, p01 is the arrival rate from good state to bad state,
which obeys the Poisson distribution. Similarly, /~10 is the
3A Adaptive Fragmentation Algorithm 4 arrival rate from b a d state to good state, am is the tran-
s i t i o n probability from good state to bad state and a m is
If(ACK lost or ACK time out) the transition probability from bad state to good state, a00
generate n E [1, v]with uniform dist.; is the transition probability from good state to good state
0k+~ = 0k + n; and a l l is the transition probability of from bad state to
if (0~+~ < ~) bad state. P9 is probability of steady-state in good state,
Ok+l = e; Pb is probability of steady-state in bad state and Perror is
Else long-term (average) probability of errors. And they follow
if(0k < ~) under equations [10]:
0k+l = 0 k x2; p s ( k + 1) = ps(k) * aoo q-pb(k) * al0
else
Ok+~ = Ok + 6; pb(k + 1) = pb(k) * alo + p b ( k ) * a l l
if (0k+l > *)
Ok+l = ~; perro, = pg * 10 -5 + Pb * 10 -2
(5)
a o l --{-aoo = 1

The algorithm 4 is designed by using the m e t h o d behind

al0 + all ---- 1
an improved version of slow-start algorithm [9]. There is a
fragment size threshold in order to slow down the increase
Pg + Pb = 1
of fragment size after this threshold is reached. ~ is the
threshold used to avoid the fragmentation size increased to In this simulation, O P N E T modeler is chosen [8]. In [1],
too large and t o o fast. issues of simulation in IEEE 802.11 W L A N using O P N E T
are discussed. But in their model, there are no noisy chan-
3.5 Adaptive Fragmentation Algorithm 5 nel and fragmentation and de-fragmentation parts of I E E E
802.11 protocol. In the simulation for this work, the time-
varying channel and fragmentation/de-fragmentation are im-
If(ACK lost or A C K time out)
plemented in order to a d a p t our requirements. The one-hop
0k+l = e;
network topology is adopted in this simulation in order to
Else
verify the link performance. In this simulation, the delay-
if(0k < ~) insensitive d a t a service traffic is used as source. Two dif-
0k+l = 0 k X 2 ;
ferent channel quality scenarios are used in order to check
else
the throughput performance of the proposed adaptive algo-
Ok+l = Ok + 6;
rithms.
if (0k+l > ~)
Ok.+l = g;
5. RESULTS AND ANALYSIS
The parameters chosen for simulations are as following, be-
cause the d a t a frame size of IEEE 802.11 is specified be-
This algorithm is more conservative in decreasing fragment tween 0 and 2312 bytes, in the simulation of this work,
size and use the same way as algorithm 4 in increasing frag- is set to 2000 bytes; the values of v and to used in algo-
ment size. rithms described before are set to 4 because of considera-

422
tions about computation complexity and real-time require- 0.5. From this figure, it is observed that under given chan-
ment, although other number could get the similar results; nel quality, using fixed fragment size, 50 bytes can give the
is set to optimal fragmentation size. Because the main con- optimal throughput.
cern here is to exam the adaptive fragmentation algorithms
in heuristic of studying congestion control algorithms im- Figure. 3 is derived from channel quality scenario I. Un-
plemented in TCP, the method of how to get optimal frame der this channel quality scenario, algorithm 1 can achieve
size adaptively is out of the reach of this paper. So far, there the best throughput performance, which is better than the
are some research working on how to get the optimal frame throughput achieved by using fixed optimal fragment size
size or fragmentation sizes [3, 5, 6]. Here, the optimal frag- (50 bytes), and all proposed algorithms using the method
mentation size for each channel quality scenario is derived behind end-to-end congestion control/avoidance algorithms
by using the method described in [5]. ~ is set to half of can give better throughput performance than that achieved
frame size, i.e., 1000 bytes. by using fixed fragment size. This is shown that when the
channel quality is getting worse, using proposed adaptive
Comparison ot'11~roughput Pmton'nance wilh Different Fb~ed Fmgmerd Size(Scenario I)
2o0o fragmentation algorithms can get more gain in terms of
throughput performance. There is more 15% increase in
16oo t ~ i i ...........
i i i l• , good throughput than the optimal throughput achieved by
i i . . . . . . . . .

i i
, . . . . . . . . . . . . . . . . . . . . .

i
. . . . . . . . . . .

i .......... ,i .......... i, ..........

,

using using fixed fragmentation size in this channel quality.

1~ ......... ] .......... i .......... ; .......... i .......... ; .......... { .......... ~.......... i ..........

CompQde, on of'll~roughput Perforrmmce with Oifle~nt Rxed Fragment S / z ~ (Scenario II)

700O ?

,~ i .......... : ! ~ . ~ , ~ . o ~ ....... ! ~,_~..,~.... ::. . . . . . i................... i . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . .......... ':.......... ! . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

SImuldon Time (mtn.) 200C . . . . . . . . . . . . . . . . . . -. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . • ................... , .................

Figure 2: C o m p a r i s o n o f T h r o u g h p u t P e r f o r m a n c e
w i t h Different Fixed F r a g m e n t Sizes (Scenario I)
i
,~eula~on "time (mn.)

Comparison of "n.lmughput Pedonmince of Adaptive Aloodthrns (Scenario I)

tsoo : F i g u r e 4: C o m p a r i s o n o f T h r o u g h p u t P e r f o r m a n c e
with Different F i x e d F r a g m e n t Sizes (Scenario II)

Clx~pattson of ThtougtN~d P e f f o ~ of Adaptive X l g o ~ , ~ g ' i . f l o II)

tloo

to 15 20 25 30 35 40 ~ 50
SlmulaUon 11me (rrdn.)

Figure 3: C o m p a r i s o n of T h r o u g h p u t P e r f o r m a n c e
: !
of Different A d a p t i v e A l g o r i t h m s (Scenario I)

15 20 25 30 35 40
~mul~on Time (INn.)

Figure.2 shows the comparison of throughput performance

using different fixed fragment sizes in channel quality sce- F i g u r e 5: C o m p a r i s o n o f T h r o u g h p u t P e r f o r m a n c e
nario I, i.e., pg (probability of steady-state in good state) o f Different A d a p t i v e A l g o r i t h m s (Scenario II)
is 0.5 and Pb (probability of steady-state in bad state) is

423
 [5] S. Ci, H. Sharif, and G. Noubir. Frame Size Analysis
Figure.4 shows the comparison of throughput performance for IEEE802.11 and Its Affect on Throughput. Proc.
u.sing different fixed fragment sizes in channel quality sce- of Sixth International PDPTA Conference
nario II, i.e.,pa is 0.7 and Pb is 0.3. From the figure, it is (PDPTA '~000), pages 2931-2937, Jun., 2000.
observed that under given channel quality, using fixed frag-
ment size, 500 bytes can give the optimal throughput. [6] E. Modiano. An Adaptive Algorithm for Optimizing
the Packet Size Used in Wireless ARQ Protocols.
Figure. 5 is derived from channel quality scenario II. Un- Wireless Networks, 5:279--286, 1999.
der this channel quality scenario, from above figures, algo- [7] G. Noubir. Multimedia Access and Distribution.
rithm 1 can achieve the best throughput performance, which European Project Internal Note, 1999.
is better than that the throughput achieved by using fixed
optimal fragment size (500 bytes). [8] OPNET Inc. OPNET Modeler User Manual. 1997.
[9] L. Perterson and B. Davie. Computer Network: A
6. SUMMARY Syster~s Approach. Morgan Kanfmann Pulishers Inc.,
In this paper, several adaptive fragmentation algorithms in- 1996.
spired by studying TCP end-to-end congestion control/avoidance
algorithms used in transport layer are proposed and studied. [10] S. Wilson. Digital Modulation and Coding.
From the above results, we observe that the proposed adap- Prentice-Hall Inc., 1996.
tive fragmentation algorithms can improve the throughput
performance of MAC layer protocol of WLAN under fad-
ing wireless channel, especially in the scenario that channel
quality is getting worse.

In general, in fading channel, the proposed adaptive frag-

mentation algorithm in heuristic of studying slow-start con-
gestion control algorithm will offer better performance in
both channel quality scenarios since the channel quality can
be reflected more accurately in this algorithm.

The proposed adaptive fragmentation algorithm proposed

here is simple and can be easily implemented. Note that al-
though the result throughput by adopting these algorithms
are better than that of an optimal fixed size, in practice
it is no optimal fixed size because we cannot have a single
static two state model, so adaptation is very important. In
future work, more efficient optimal fragmentation size esti-
mation methods and more complex channel models should
be considered.

7. ACKNOWLEDGEMENTS
The authors of this paper would like to thank two anony-
mous reviewers for their valuable comments to help us to
carefully make the presentation.

8. REFERENCES
[1] R. Baldwin. IEEE80~.11 Wireless LAN Model
Documentatioa OPNET modeler, 1998.

[2] H. Chhaya and S. Gupta. Performance Modeling of

Asynchoronous Data Transfer Methods of IEEE 802.11
MAC Protocol. Wireless Networks, 3:217-234, 1997.

[3] C. Chien, M. Srivastava, R. Jain, P. Lettieri,

V. Aggarwal, and R. Sternowski. Adaptive Radio for
Multimedia Wireless Links. I E E E Journal on Selected
Areas in Communications, 17(5):793-813, 1999.

[4] S. Ci and H. Sharif. Adaptive Approaches to Enhance

Throughput of IEEE 802.11 Wireless LAN with
Bursty Channel. Proc. of The ~5th Annual IEEE
Conference on Local Computer Networks (LCN ~000),
Nov., 2000.

424

View publication stats