1. Enhancement of IEEE 802.11 Distributed
Coordination Function with Exponential Increase
Exponential Decrease Backoff Algorithm
Nai-Oak Song, ByungJae Kwak, Jabin Song, Leonard E. Miller
Advanced Network Technologies Division
National Institute of Standards and Technology, MD, USA
Emial:{nsong, bjkwak, jabin, Imiller} @antd.nist.gov
Abstract- A new backoff algorithm is proposed to enhance
the performance of the IEEE 802.11 Distributed Caordination
Function (DCF) which employs Binary Exponential Backoff
(BEB) algorithm. The proposed algorithm, called the Exponen-
tial Increase Exponential Decrease (EIED) backoffalgorithm, is
quite simple to implement while significantlyimproving the net-
work performance over BEB. Another hackoff algorithm d e d
Multiple Increase Linear Decrease (MILD) backoff algorithm is S:ruccerr Ciollirim
considered for performance comparison. The simulation results
show that ElED outperforms BEB and MILD in terms of both ~ i I. ~ Backoff mechanism of BEB: CW,;.
. = 16. CW,, = 1024.
throughput and delay. The performance gain of ElED comes m = 7.
from successfully balancing the two extreme backoff policies of
BEB and MILD.
lnder Tcms- Wireless LAN, DCF, Binary Exponential Back- to the minimum after a successful transmission, or when the
off, EIED, MILD. total number of transmission attempts for a packet reaches
the limit m (m = 7 for basic access mechanism and
1. INTRODUCTION m = 4 for the Request-To-SeluVClear-To-Send (RTS/CTS)
Combined with wireless communication technologies such exchange mechanism). However, the contention window
as Wireless LAN (IEEE 802.1 I), Bluetooth, and HIPERLAN, resetting mechanism causes a very large variation of the
powerful notebook computers and PDAs have created a new contention window size, and degrades the performance of
mobile computing environment. Because of its already well a network when it is heavily loaded since each new packet
established market acceptance, IEEE 802.1 I is the most suc- starts with the minimum contention window, which can he too
cessful among the above mentioned wireless communication small for the heavy network load. Fig. I illustrates the backoff
standards. The Distributed Coordination Function (DCF) is mechanism of BEB, where CW,,,j. = 16 and CW,, = 1024
the fundamental access mechanism in IEEE 802.1 I Medium (m = 7). To resolve the problem of BEB, 121 proposed a
Access Control (MAC) while the Point Coordination Func- backoff algorithm known as MILD (Multiple Increase Linear
tion (PCF) used optionally [I].In DCE Binary Exponential Decrease), where the contention window size is multiplied
is
Backoff (BEB)’ is employed as a stability strategy to share by 1.5 on a collision but decreased by 1 on a successful
the medium. At the first transmission attempt of a packet, transmission as follows:
BEB selects a random slot from the next CW = CW,i, CW = min[l.5. CW, CW,,,] on a collision, (2)
slots with equal probability for transmission, where CWmin
is the minimum contention,window size. Every time a node’s CW = max[CW - 1,CW,,i.] on a success. (3)
packet is involved in a collision, the contention window size MILD performs well when the network load is steadily heavy.
for that node is doubled up to its maximum CW,, that is, However, this extremely conservative transmission policy has
CW = min[2 . CW, CW,,] (1) its shortcomings: MILD does not perform well when the
network load is light because it takes quite long time to
and the new contention window is used for the following recover from the backoff caused by occasional collisions.
transmission attempt. A node resets its contention window Furthermore, when the number of active nodes changes
‘In IEEE 802.1 I. a tnincaled BEB is used. Many papen refer to this sharply from high to low, MILD cannot adjust its CW
truncated version as BEB. We follow the convention in this paper. fast enough because its “linear decrease” mechanism. For
0-7803-7757-51031$17.~OZW3 lF€E 2775

2. example. when CW,i, = 16 and CW,, = 1024 (IEEE
802.1 I using Frequency-Hopping spread spectrum (FHSS)
physical layer (PHY)). it takes a maximum of 1008successful
transmissions for MILD to reach CW,i,. Another extreme is
BEB, which takes only one successful transmission to reach
CWmi,.
There are also other contention resolution mechanisms
Ccollirioo
S.(UCCC~S
proposed to improve the network performance of IEEE
802.1 I DCF. However, most of them require exchange of Fig. 2. Backoff mechanism of EIED: CW.,jn = 16. C
W
,, = 1024.
information between nodes and complicated computation [3], I = 2. T D = A.
,
[4], [ 5 ] . Those algorithms are not considered and we consider
only acknowledgment based backoff algorithms [6]. In this
paper, an exponential increase exponential decrease (EIED) InterFrame Space (EIFS). Since the number of transmission
backoff algorithm is proposed to enhance the the performance retries is bounded by m, a packet must be dropped after m
of the IEEE 802.1 I DCF. EIED is as simple as BEB to transmission retries, that is, after experiencing m times of
implement while significantly improving the performance of packet collision.
IEEE 802. I I DCF,
This paper is organized as follows. First, we briefly de- 111. EIED, THE PROPOSED ALGORITHM
scribe the DCF of the IEEE 802.1 l MAC in Section 11. The As explained in Section 11, the IEEE 802.1 I DCF employs
proposed backoff algorithm. EIED, is presented in Section BEB as a stability strategy to share the medium. But its
111. Then, in Section IV the performance of EIED is evaluated contention window resetting mechanism degrades the perfor-
through simulation and compared with BEB and MILD. mance of a network. In this section, we propose EIED backoff
Finally, we conclude this paper in Section V. algorithm, where the contention window size is increased and
decreased exponentially on collision and successful transmis-
11. IEEE 802.1 I DCF sion, respectively, to enhance the performance of the IEEE
DCF is the fundamental access mechanism in the IEEE 802.11 DCE
802.1 I MAC. In this section we will briefly describe the In EIED, whenever a packet transmitted from a node is
basic access mechanism of DCF. involved in a collision, the contention window size for the
When a node (station) receives a packet to transmit, if node is increased by backoff factor rr. and the contention
the node senses the medium has been idle for a period window for the node is decreased by backoff factor PD if
of time longer than or equal to a Distributed InterFrame the node transmits a packet successfully. The EIED backoff
Space (DIFS), the packet transmission may begin at the algorithm can be represented as follows.
beginning of the immediately following slot. Otherwise, the
node should defer the packet transmission as follows. The CW = min[rr .CW,CW,,] on a collision,
node waits until the medium is idle for a DIFS and then CW = max[CW/rD,CW,i,] on a success.
sets its backoff timer. The backoff timer is set to a value
which is randomly selected fmm 0,1,. .- ,CW - 1 with As shown in the simulation results below, the performance
equal probability, where CW represents contention window of EIED is affected by the choice of the values of and
T D . In this paper, we fully exploit the degrees of freedom
size. Initially, contention window size is set to its minimum
of EIED to obtain a better performance. Fig. 2 illustrates the
CW,i. for the first transmission attempt of a packet. Every
time the packet is involved in a collision, contention window backoff mechanism of EIED with rr = 2 and r D = d. A
size is doubled for the next transmission attempt. However, special case of our proposed scheme with rr = rD = 2 was
the contention window size cannot exceed its maximum presented in (71,where the throughput was compared with
CW,,. The backoff timer is decreased by I evety slot time BEB in IEEE 802.1I DCF under saturation condition. Note
that it takes maximum 12 successful transmission for EIED
(50 psec for FHSS PHY) after the medium has been idle for
a DIFS, but is frozen when the medium becomes busy, When with T r = 2 and T D = fi to reach CW,,. in the above
the backoff timer becomes zero, the station transmits the example.
packet. If the destination node receives the packet correctly, it
sends a positive acknowledgment (ACK) packet after a Shori IV. SIMULATION
InterFrame Space (SIFS). When the source node does not The Wireless LAN model of the network simulator OP-
receive an ACK, it assumes the packet has experienced a col- NET 8.1.A was used for the performance comparison of
lision and updates the contention window size CW according BEB, MILD, and EIED. The normalized throughput with
to BEB algorithm as described above, then sets its backoff respect to the channel capacity, and delay defined as the
timer to a newly selected backoff values after a Extended time from the moment a packet is placed in a queue until
2776

3. (b) Delay
Fig. 3. Simulation results: normalized throughput and delay with respect to padel mival rate [packetslssec]. N = 5,10,40,60.
2777

4. the beginning of its successful transmission were used as the V. CONCLUSION
performance we the performance when In this paper, we proposed a new backoff algorithm (EIED)
the basic access mechanism with FHSS PHY was used, where enhance the performmce of IEEE 802.11 DCF, The
CW,i, = 16, CW,, = 1024. m = 7. Fig. 3 shows the performance of EIED was compared with BEB and MILD
simulation results for number of nodes N = 5, 10.40, and 60 using O ~ N E Tnetwork simu~ator, ne simulation results
for various system wide packet arrival rates from 10 to 160 show that EIED outperforms BEB and MILD in terms of both
packetdsec. Thus the number of nodes combined with the throughput and delay. The performance gain of comes
packet arrival rate is defined as the offered load. w e assumed from successfully balancing the two extreme backoff policies
that packets =e 1024 bytes long and the arrival process of BEB and MILD. BEB does not use the collision history
is Poisson. BEB and MILD are compared with four different of the previous packe& and reduces cw too fast making it
cases of EIED: not suitable for network under heavy load (large N ) . On the
(1) rr = 2, T D = z'/* other hand, MILD'S linear decrease policy is too conservative
(2) = 2, TD = zil4 especially when N is small. EIED is highly customizable by
(3) T I = T D = 2 f i parameters TI and TD. Simulation results show that EIED
(4) T I = T D = 2 with relatively smaller value of T~ compared to the value of
" has higher performance gain'
The figure shows that when the packet arrival rate is low
(approximately < 80 packetdsec), all three backoff mecha-
nisms have almost the same throughput which is proportional REFERENCES
to the arrival rate, regardless of the number of nodes. The [ I ] IEEE 802 P a n 11: Wirrlvsr U N M e d i u r n Access Conlml (MAC) m d
four cases of EIED always give higher throughput than BEB. Phyriral Lover (PHYj sprci/iratiom. IEEE Std.. 1999.
[21 v, Bhargha"an, A. Demen, s. she*r. and L, Zhang. "MAC-w:
When the number of nodes is small, the performance of BEB A ACCO~~ pmtaoI for Wireless LAN's: in ~ m c ACM SIG-
.
is almost as good as that of EIED, but as the number of COMM'94. London. England, 1 9 . pp. 212-225.
94
nodes increases the performance differences also increase, 131 G. Bianchi. L. Fratta. and M. Oliveri. "Pelfomance Evaluation and
Enhancement of the CSMNCA MAC Pmtaol for 8M.11 Wireless
As a result, the throughput of BEB iS Only about 213 Of that LAN~:' in pmr. piMRC'96. Taipei. ~ a i w a n Oct. 19%. pp. 392-3%.
.
of the first two cases of EIED for N = 60. The throughput 141 F. Cali. M. Conti. and E. Fregori. "Dynamic Tuning ofthe IEEE 802.11
of MILD is very low when the network load is light ( N = 5 ) F'rolaol to Achieve a Theoretical Throughput Limil," IEEWACM T h s .
Nerworking. vol. 8. no. 6. pp. 785-799. Dec. 2000.
as expected. However. it becomes higher as the number of [SI D. Qiao and K. G.Shin. "UMAV: A Simple Enhancement to the EEE
nodes increases and its performance is on a par with EIED 802.11 DCF." in PNC Ihr 36lh Hawaii lnamnrio,ml CoBferencr on
with " = 2 ' T D = 211a for = 40' For even higher (HICSS-36). Hawaii. Jan. 2003.
Sy.mm S c i m c ~ s
161 F. P. Kelly and I. M. McPhee. T h e Number of Packets Transmilled
number of nodes ( N = 60). MILD yields only slightly higher by Collision Detect Random A C C ~ S SSchemes," Annals of Pmbabiliry.
throughput than Elm. vol. IS. pp. 1557-1568. 1987.
As in the case of throughput, for all N , all EIEDs have 171 H. WU. s. Chew. y Pen& K. L O W and J . Ma. ' K E E 802.11
.
Dist6baed Cwrdination Function (DCF): Analysis and Enhancement:'
smaller delays than BEB for a[[arrival rates, and outperforms in pmC. I C C Z ~N~~ yo*. 2 ~ pp, .605409,
~ .
MILD for most of the arrival rates. The delay performance
of MILD is especially poor when the number of nodes is
small. The poor delay performances of BEB and MILD occur
for different reasons. The delay performance of BEB suffers
from packet collisions which cause more retransmissions.
Thus, the delay performance of BEB is worse when there are
more contending nodes (more collisions). On the other hand,
the delay performance of MILD suffers from the excessive
backoff, and thus the performance of MILD is worse when
there are smaller number of nodes where even occasional
collisions could c a w unnecessarily large backoff.
2778