1
Undesirable service differentiation in future WLANs Srikant Kuppa, Shashidhar Rao Gandham and Ravi Prakash Department of Computer Science The University of Texas at Dallas Richardson, TX 75083 Email: ksrikant, gshashi, ravip @utdallas.edu
Abstract The IEEE 802.11 Distributed Coordination Function DCF offers only best-effort service. The Enhanced DCF EDCF scheme supports quality of service by establishing a probabilistic priority mechanism to access the shared wireless medium. four
access namely, , EDCF , defines and categories, to support voice, video, best-effort background traffic, respectively. Since and DCF and of EDCF offer best-effort service, it is desirable that applications demanding such a service, experience comparable delay and throughput whether they are run on a DCF-compliant or an EDCF-compliant wireless station. In this paper, we show through simulation experiments that DCFand EDCF-compliant stations do not provide comparable support to best-effort applications while operating together. This is due to different parameter settings in DCF and EDCF which leads to an undesirable service differentiation.
1. Introduction In IEEE 802.11 Distributed Coordination Function (DCF) [1], wireless stations maintain a backoff counter, which is initialized in the range "!$# % &('*),+.-*# % &('0/2134 , where 5 is the number of retransmission attempt(s) associated with the head-of-line frame, and #(% & '*),+ and #(% & '0/21 refer to the minimum and maximum contention window size. During backoff procedure, stations sense the wireless medium. If the medium is sensed idle during a slot, then backoff counter is decremented by one. Otherwise, the value of backoff counter is frozen and the backoff procedure is re-initiated only after sensing the medium idle for distributed inter-frame space (DIFS). Stations are permitted to transmit only when their backoff counter reaches zero. DCF scheme does not prioritize access to the wireless medium. The upcoming IEEE 802.11e [5] Enhanced DCF (EDCF) scheme coordinates access to the shared wireless medium based on traffic priorities by establishing a probabilistic priority mechanism. access (ACs), , EDCF defines , four and categories
to support namely, voice, video, best-effort and background applications, respectively. These ACs independently access the shared wireless medium with parameters like arbitration inter-frame spaces (AIFS) and contention window (CW) sizes. Higher the AC, smaller the values of AIFS and/or CW sizes.
The only difference between of EDCF and DCF lies in the number of contiguous slots that need to be sensed idle prior to (re-)initiation of the backoff procedure. The number of such contiguous slots in DCF and of EDCF are 687:9<;>=?;@7:9<;BAC D!E#F;HG(IKJFL8MON8P and Q87R9H;>= ;@7R9H;SAUTV!W#F;FG(I JFL8M3N8P , respectively. Here, SIFS refers to short inter-frame space and aSlotTime denotes the duration of a slot. Thus, there is a difference of one slot in the inter of frame space (IFS) values used by DCF and EDCF. In [6] and [7], the authors have reported that even a small difference in the values of IFS leads to substantial service differentiation. Thus, in our particular case, it appears that best-effort applications running on DCF- and EDCF-compliant stations may experience differentiated service, an outcome that is highly undesirable. To the best of our knowledge, there is no study in the literature that examines the interoperability between DCF and EDCF. Such a study bears great significance because once EDCF-enabled wireless cards are released into the market, they may have to operate along with DCF-enabled wireless cards in the same frequency band [4]. Previous studies have focused on WLANs operating under saturation conditions, that is, when all stations in the network have data to transmit at all times. It is not clear whether the assertions made in the earlier studies are valid under all load conditions. Also, do the previous results apply specifically when DCF and of EDCF operate together, which is a realistic scenario to occur in the near future? Thus, it is to be verified whether the disparity to best-effort traffic occur between DCF with ofrespect and EDCF: (i) under different load conditions, and (ii) irrespective of the PHY layer characteristics. The problem at hand can be stated as follows. Consider a one-hop ad hoc network with XZY3[2\ DCF-compliant stations and XS]^Y3[2\ EDCF-compliant stations. It is desired that besteffort applications running on DCF- and EDCF-compliant stations experience comparable delay and throughput under the following three network scenarios: (i) X Y3[2\S_ V` X ]^Y3[2\ =a , (ii) X Y3[2\ =ab`cX ]^Y3[2\S_ , and (iii) X Y3[2\S_ b`cX ]^Y3[2\S_ . The objective of our work is to verify whether such a desirable outcome is realizable or not. The rest of this paper is organized as follows. We describe the simulation model in Section 2. In Section 3, we validate the differences in support offered to best-effort traffic using simulation results. Finally, Section 4 concludes this paper.
2
2. Simulation model
3. Simulation results and discussion
We used the implementation of the basic channel access mechanism of DCF and EDCF schemes in Network Simulator H (NS- H ) tool [2] for simulations. System parameters of EDCF were initialized to the following values for (as per Table K< on page in [5]): (i) Q87:9<; =U;@7:9<; A T! #(;HG(I JFL MON8P , (ii) minimum CW size =E #(% &('3),+ , (iii) maximum $ = . System CW size = #(% & '0/21 , and (iv) parameters like channel bandwidth ( ), Slot time (aSlotTime), Short IFS (SIFS), Distributed IFS (DIFS), minimum contention window ( #(% & '*),+ ) and maximum contention window ( #(% & '0/21 ) were assigned based on the PHY layer specification.
Since DCF and EDCF are MAC sublayer specifications, and both DCF- and EDCF-compliant stations would operate in the same frequency band [4], we conducted simulations using both IEEE 802.11b [1] and IEEE 802.11a PHY layer specifications [3]. Simulation results with 802.11a PHY can be found in [8], which is an extended version of this paper. Three different physical layer specifications are defined in the IEEE 802.11b standard [1]. In this paper, we considered the widelyused direct sequence spread spectrum (DSSS) specification which uses #(% & '*) + =bT( and #(% & '*),+ =L: M .
2.1. Network scenario
We considered three network scenarios with: (i) fixed XcY3[2\ and varying XS]4Y3[2\ , (ii) fixed XS]^Y3[2\ and varying XSY3[2\ , and finally, (iii) varying both X Y3[2\ and X ]4Y3[2\ keeping X Y3[2\ A X ]^Y3[2\ fixed. In all the three network scenarios, the stations were kept stationary and within range of each other forming a one-hop ad hoc network. All the stations communicated with a single dummy station. For simplicity, we assumed the wireless medium to be free from capture effect and bit errors.
2.2. Traffic scenario All stations generated traffic demanding best-effort service with frame size equal to R . We constructed two traffic scenarios to operate the network under saturation and nonsaturation conditions, respectively. Suppose the load offered by the best-effort running on all the wireless stations
.applications is % of Then, each station generated traffic according "!$#&% to CBR distribution at the rate of ')(+*-,/.0'&1-(+* ,32 54 . The value of was set to for simulating saturation conditions. In order to investigate the disparity between DCF and of EDCF under more realistic traffic scenarios, non-saturation conditions were simulated. Different values of used for simulating non-saturation condition were , . and 6 . All the simulation experiments were run for 798 0: of simulated time. In order to observe steady state behavior of the stations and eliminate transient measurements during the warm-up period, data obtained from the first 798 0: of each simulation run were not taken into account1.
2.3. Performance metrics Latency (in seconds): Mean delay experienced by a data frame transmitted by a station. Throughput (in Mbps): Number of data bits successfully transmitted per unit time. We measured: (i) throughput per station, (ii) cumulative throughput over all DCF- and EDCFcompliant stations, and finally, (iii) network-wide throughput. Let ; Y3[2\ and :; ]^Y3[2\ be the respective throughputs of a DCF- and EDCF-compliant station. Then, cumulative throughput over all DCF-compliant stations is computed as XZY3[2\ !<; Y3[2\ . Similarly, cumulative throughput over all EDCF-compliant stations is X ]^Y3[2\!=; ]^Y3[2\ . Finally, network-wide throughput is computed as X Y3[2\ !>; Y3[2\ AWX ]^Y3[2\ !?; ]^Y3[2\ . 1 In our simulations, we made sure that all the stations acquired reachability information and populated their ARP caches well before @BADCFEFGBHFIJKC .
Analysis under saturation conditions We now present the simulation results obtained by operating the network consisting of both DCF- and EDCF-compliant stations under saturation conditions. Fixed number of DCF-compliant stations Fig. 1 depicts the results obtained by running simulations with fixed number of DCF-compliant stations. We observe the disparity between the throughput curves of each DCFcompliant (:; Y3[2\ ) and EDCF-compliant (;:]4Y3[2\ ) station. Ideally, we would want these two curves to overlap. The disparity occurs due to different IFS which is the only difference settings, of EDCF. between DCF and Also, we observe that the delay trends are in correspondence with throughput curves. It is interesting to note that the network-wide throughput 2 54 irrespective of the reaches the saturation limit of distribution of number of DCF- and EDCF-compliant stations. Fixed number of EDCF-compliant stations In fig. 2, we once again observe the disparity between the throughput curves for DCF-compliant and EDCF-compliant stations. Let equivalence point be defined as the point of intersection between the curves of cumulative throughput over all DCF-compliant and cumulative throughput over all EDCFcompliant stations. At equivalence point, XZY3[2\$!N; Y3[2\ = X ]4Y3[2\ !>; ]^Y3[2\ . In fig. 2, X Y3[2\PO and X ]^Y3[2\ =Q at equivalence point, meaning cumulative throughput of five DCFcompliant stations is equivalent to the cumulative throughput of ten EDCF-compliant stations. This means that relatively fewer number of DCF-compliant stations are able to hog the shared channel, despite the presence of a large number of EDCF-compliant stations. Fixed number of stations in the network Fig. 3 depicts the results obtained by running simulations with fixed number of stations in the network and varying the proportion of DCF- and EDCF-compliant stations. Significant observations related to this simulation run are: R At equivalence point, XSY3[2\ OTS and X ]^Y3[2\ =E K . R :; Y3[2\ _ ; ]^Y3[2\VU Under heavy loads, DCF-compliant stations consistently win over their EDCF counterparts. R ; Y3[2\ in a network with X Y3[2\ =W>(X 0: X ]^Y3[2\ = MY _ ; ]^Y3[2\ in a network with X Y3[2\ =E YZ(X 0 : X ]^Y3[2\ =[ . This is because IFS value used by DCF-compliant stations is one slot less than the IFS value used by EDCF-
(a) Ndcf=10
1
Throughput (in Mbps)
10
0
10
0
2
4
6
8
10
Mean frame exchange delay (in seconds)
3
(b) Ndcf=10 0.05 Per station (DCF) Per station (EDCF)
0.04 0.03 0.02 0.01 0
0
2
4
N
6
8
10
N
edcf
edcf
Per station (DCF) Per station (EDCF) Cumulative (DCF) Cumulative (EDCF) Network−wide
Throughput and delay variations by fixing number of DCF-compliant stations (a) Nedcf=10
1
Throughput (in Mbps)
10
Equivalence point 0
10
0
2
4
6
8
10
Mean frame exchange delay (in seconds)
Fig. 1.
(b) Nedcf=10 0.05 Per station (DCF) Per station (EDCF)
0.04 0.03 0.02 0.01 0
0
2
4
N
6
8
10
N
dcf
dcf
Per station (DCF) Per station (EDCF) Cumulative (DCF) Cumulative (EDCF) Network−wide
Throughput and delay variations by fixing number of EDCF-compliant stations (a) Ndcf+Nedcf=30
1
Throughput (in Mbps)
10
Equivalence point
0
10
0
10
20
30
N
dcf
Mean frame exchange delay (in seconds)
Fig. 2.
(b) Ndcf+Nedcf=30 0.10 Per station (DCF) Per station (EDCF)
0.08 0.06 0.04 0.02 0
0
10
20
30
N
dcf
Per station (DCF) Per station (EDCF) Cumulative (DCF) Cumulative (EDCF) Network−wide
Fig. 3.
Throughput and delay variations by fixing total number of stations
compliant stations. This makes DCF-compliant stations more aggressive. R ; Y3[2\ O ; ]^Y3[2\ in a network with X Y3[2\ =a YZX(0: X ]^Y3[2\ = . By using smaller IFS values, and thereby, being more aggressive, DCF-compliant stations experience more collisions. This reduces per DCF-station throughput (; Y3[2\ ). Analysis under non-saturation conditions Figs. 4-6 illustrate the results obtained by operating the network under non-saturation conditions. The disparity between the curves of ;:Y3[2\ and ; ]^Y3[2\ increases with increase in load (i.e. value of ). As the load increases, DCF-compliant stations manage to get higher bandwidth share by being more aggressive. So, fewer the number of more aggressive DCFcompliant stations, lesser is the contention among them and greater is their ability to hog the channel. This is illustrated in
figs. 4(c), 5(c) and 6(c), wherein we observe that greater the value of , smaller the value of X Y3[2\ at equivalence point. Based on our simulation results, we infer that DCF- and EDCF-compliant stations do not provide comparable support to best-effort applications while operating together. More importantly, different IFS settings in DCF and EDCF results in an undesirable service differentiation even though all the stations are running only best-effort applications.
4. Conclusions and future work In this paper, we examined the interoperability between DCF and EDCF by considering only best-effort applications. It is very important that these applications experience comparable delay and throughput whether they are run on a DCFcompliant or an EDCF-compliant station. Otherwise, there
4
(a) Ndcf=10
1
(b) Nedcf=10
1
10
0
10
0
10
0
10
(c) Ndcf+Nedcf=30
1
10
10
Equivalence point −1
10 0
5
10
0
5 Ndcf
Nedcf
10
0
10
Per station (DCF) Per station (EDCF) Cumulative (DCF) Cumulative (EDCF) Network−wide
Fig. 4. (a) N
Throughput (in Mbps) analysis under non-saturation conditions when =10
(b) N
dcf
1
=10
Equivalence point
10
0
10
0
5 Nedcf
10
0
5 Ndcf
10
0
10
(a) N
Throughput (in Mbps) analysis under non-saturation conditions when =10
(b) N
dcf
1
=10
Equivalence point
10
0
10
0
5
10
0
5 N
N
edcf
10
0
10
5. References [1] IEEE Standard for Wireless LAN Medium Access Control (MAC) and Physical (PHY) Layer Specifications, ANSI/IEEE Std 802.11, 1999 Edition. [2] Network Simulator, Webpage - http://www.isi.edu/nsnam/ns/. [3] Supplement to Part 11 standard for Wireless LAN Medium Access Control (MAC) and Physical (PHY) Layer Specifications - High speed physical layer in the 5 GHz band, ANSI/IEEE Std 802.11a-1999.
30
dcf
Throughput (in Mbps) analysis under non-saturation conditions when
might be hesitation of some users to upgrade to EDCF. onofpartEDCF DCF and differ only in IFS settings. In this paper, we have shown that this seemingly small difference leads to substantial disparity in treatment offered to best-effort applications. Such a disparity occurs irrespective of the PHY layer specification. Also, under heavy loads, DCF-compliant stations consistently win over their EDCF counterparts, an outcome that is highly undesirable.
20 N
dcf
Per station (DCF) Per station (EDCF) Cumulative (DCF) Cumulative (EDCF) Network−wide
Fig. 6.
=30
edcf
10
0
0
+N
dcf
1
10
10
30
A
(c) N
edcf
1
10
20 Ndcf
Per station (DCF) Per station (EDCF) Cumulative (DCF) Cumulative (EDCF) Network−wide
Fig. 5.
=30
edcf
10
0
0
+N
dcf
1
10
10
30
(c) N
edcf
1
10
20 Ndcf
[4] Avaya Wireless 5-GHz Educational Primer, Webpage http://www.essentia.it/documenti/WP Articoli Tecnici/Avaya-Wireless 5GHz Educational Primer.pdf, October 2001. [5] Draft supplement to Part 11 - Wireless Medium Access Control (MAC) and Physical (PHY) Layer Specifications: MAC enhancements for Quality of Service (QoS), IEEE P802.11e/D8.0, February 2004. [6] Chun-Ting Chou, Kang G. Shin and Sai Shankar N. Inter-Frame Space (IFS) based service differentiation for IEEE 802.11 Wireless LANs. In Vehicular Technology Conference, pages 1412–1416, October 2003. [7] Giuseppe Bianchi and Ilenia Tinnirello. Analysis of priority mechanisms based on differentiated inter-frame spacing in CSMA/CA. In Vehicular Technology Conference, pages 1401–1405, October 2003. [8] Srikant Kuppa, Shashidhar Rao Gandham and Ravi Prakash. Undesirable service differentiation in future WLANs. In UTD Computer Science Tech. Report, UTDCS-30-04, http://www.utdallas.edu/ ksrikant/publications.htm, August 2004.
