- Email: [email protected]
Modified EDCF to improve the performance of IEEE 802.11e WLAN
Computer Communications 30 (2007) 841–848 www.elsevier.com/locate/comcom
Modified EDCF to improve the performance of IEEE 802.11e WLAN Wen-Yen Lin
a,b
, Jung-Shyr Wu
c,*
a
c
Department of Electrical Engineering, National Central University, Taiwan, ROC b Department of Electronic Engineering, Vanung University, Taiwan, ROC Department of Communication Engineering, National Central University, Taiwan, ROC
Received 22 December 2005; received in revised form 11 October 2006; accepted 19 October 2006 Available online 21 November 2006
Abstract In this paper, we propose a modified EDCF scheme, M-EDCF, to improve the Quality of Service (QoS) of the IEEE 802.11e wireless network. The IEEE 802.11e standard is presented to support QoS at medium access control level using a priority scheme by differentiating the inter-frame space and the initial window size. In addition to providing relative priorities by adjusting the size of the Contention Window (CW) of each traffic class, our proposed scheme, M-EDCF, also consider the effect of a back_off_timer to avoid unnecessary collisions. Our study shows that in either in heavy or light traffic load our proposed scheme can provide better quality for both high priority and low priority packets than either the AEDCF [L. Romdhani, Ni. Qiang, T. Turletti, Adaptive EDCF: enhanced service differentiation for IEEE 802.11 wireless ad-hoc networks, IEEE Wireless Communications and Networking, Conference. vol. 2, March (2003) pp. 16–20] or the original EDCF. Ó 2006 Elsevier B.V. All rights reserved. Keywords: 802.11e; Enhance distributed coordination function (EDCF); Quality of service (QoS); Wireless LAN; 802.11
1. Introduction The IEEE 802.11 enables fast installation, with minimum management and maintenance costs, and is a very robust protocol for the best-effort service in the wireless medium. However, it is unsuitable for multimedia applications with Quality of Service (QoS) requirements. The IEEE 802.11 medium access control (MAC) [2] employs a mandatory contention-based channel access function called distributed coordination function (DCF), and an optional centrally controlled channel access function called point coordination functions (PCF). The DCF adopts a carrier sense multiple access with collision avoidance (CSMA/ CA) with binary exponential backoff. However it does not support QoS, and cannot support any priority scheme even through the PCF can support very limited QoS, as an
*
Corresponding author. Tel.: +886 913976255; fax: +886 34514141. E-mail addresses: [email protected], [email protected] (J.-S. Wu).
0140-3664/$ - see front matter Ó 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.comcom.2006.10.013
optional function that has not been implemented in the most of current products. Since its introduction, various extensions and modifications have been studied to address this current need and the IEEE 802.11 Task Group E [3] is responsible for developing a QoS-aware MAC protocol that considers several service differentiation mechanisms. However, the performance of service differentiation has only been evaluated for a supplement to the original IEEE 802.11 MAC. The IEEE 802.11e MAC will support multimedia applications such as voice and video over the IEEE 802.11 WLAN. The IEEE 802.11e MAC also employs a contentionbased channel access function called the enhanced distributed coordination function (EDCF), and a centrally controlled channel access function called the Hybrid Coordination Function (HCF). The EDCF provides a priority scheme by differentiating the inter-frame space and the initial window size. Such a priority scheme may provide prioritized but not guaranteed QoS due to it’s contention-based nature. Prioritized QoS will be useful for those
842
W.-Y. Lin, J.-S. Wu / Computer Communications 30 (2007) 841–848
multimedia applications that do not require rigid QoS. One advantage of prioritized QoS is that it is simple to implement since it looks like the DiffServ model. In this article, we focus on the EDCF, but not the HCF. Quality of Service (QoS) support is critical to multimedia applications. Here, time-bounded services such as audio and video conference typically require some specified bandwidth, delay and jitter guarantee, but can tolerate some losses. However, in DCF all the stations in a Basic Service Set or all the flows from the same station compete for resources and channels with the same priority. There is no differentiation mechanism to guarantee packet delay and jitter to stations or flows to supporting time-bounded multimedia services. Several priority studies have been reported in the literature for DCF and EDCF, and priority schemes can be classified into three kinds: Backoff based priority schemes [9–11], IFS-based priority schemes [12,13], and hybrid priority schemes [1,6–8,14–16]. The hybrid schemes, such as the EDCF [3], adopt both a backoff based scheme and an IFS-based scheme, and most recent reports focus on it [1,6–8]. Although those mechanisms improve the quality of service of real-time traffic, the performance levels obtained are not optimal since EDCF parameters cannot be adapted to the network conditions. When the collision rate increases very fast, the contentions to access the shared medium are very high, which significantly affects the throughput. This motivates us to propose a scheme that adapts the CW parameter and back_off_timer according to the network conditions. Our proposed scheme is called the M-EDCF, and is a hybrid scheme based on Ref. [1] to improve the performance of throughput, mean delay and dropping rate in the buffer. In this paper, we introduce the 802.11e EDCF priority scheme and AEDCF in Section 2. In Section 3, we describe our proposed scheme, called the modified AEDCF priority (MA_EDCF) scheme. Simulation results are studied in Section 4. Section 5 concludes the paper by summarizing results and outlining future research.
2. IEEE 802.11e and adaptive EDCF The 802.11 legacy MAC does not support the concept of differentiating frames with different priorities. Basically, the DCF is supposed to provide channel access with equal probabilities to all stations contending for the channel access in a distributed manner. However, equal access probabilities are not desirable among stations with different priority frames. The emerging EDCF is designed to provide differentiated and distributed channel access for frames with eight different priorities (from 0 to 7) by enhancing the DCF, as shown in Fig. 1. In contrast to original DCF, the EDCF is not a separate coordination function. Rather, it is a part of a single coordination function, called the Hybrid Coordination Function (HCF), of the 802.11e MAC. The HCF combines the aspects of both DCF and PCF. All the detailed aspects of the HCF are beyond the scope of this paper since we focus on the HCF contention-based channel access, i.e., EDCF. Each frame from the higher layer arrives at the MAC along with a specific priority value. Then each QoS data frame carries its priority value in the MAC frame header. An 802.11e STA implements four access categories (ACs), where an AC is an enhanced variant of the DCF 0. Each frame arriving at the MAC with a priority is mapped into an AC as shown in Table 1. Table 1 Category mappings of priority to access Priority
Access category (AC)
Designation (informative)
1 2 0 3 4 5 6 7
0 0 0 1 2 2 3 3
Best effort Best effort Best effort Video probe Video Video Video Video
Fig. 1. Basic DCF vs queue-based EDCF.
W.-Y. Lin, J.-S. Wu / Computer Communications 30 (2007) 841–848
2.1. HCCA (polling-based-HCF channel access scheme) The HCCA is similar to the channel access scheme of PCF in that both are served by AP. When using HCCA, the AP can obtain the control of wireless channel after a time interval of PIFS, and start up the working station in the polling area, where information needs to be delivered. Before this, the AP must obtain traffic flow information for each working station through controlling frames so that AP can use this information to arrange the wireless sources. As in PCF, HCCA can also establish the Superframe, which is composed of CP-contention phase and CFP-contention free phase. At the same time, the AP is allowed to initiate delivery at any time when information needs to be delivered. 2.2. EDCA (contention-based-EDCF) Basically, an AC uses AIFS[AC], CWmin[AC], and CWmax[AC] instead of DIFS, CWmin, and CWmax, of the DCF, respectively, for the contention process to transmit a frame belonging to access category AC, as shown in Fig. 2.The AIFS[AC] is determined by AIFS½AC ¼ SIFS þ AIFS½AC SlotTime
ð1Þ
where AIFS[AC] is an integer greater than zero. Moreover, the backoff counter is selected from [1, 1+CW[AC]]. 2.3. Related improved schemes for the IEEE 802.11 To date, there are many studies on 802.11 wireless networks supplied with instant information as quality guarantee. They are classified into two groups, one is station-based improvement [11,12], and the other one is queue-based improvement [22]. The former indicates represents that each working station has its own special parameter, and the latter represents that each working station has many queues, which serve as simulated working stations, and each queue has its own parameter. 2.4. AEDCF (adaptive EDCF) illustration After each successful delivery, the EDCF will reset its contention window to CWmin, regardless of present
Fig. 2. The EDCF channel access.
843
network conditions. However, after a collision occurs, the chance of having the second collision will be higher within a short time, so the method mentioned in [1] is based on the network condition, and gradually lessens the contention window instead of resetting the windows value to CWmin directly to avoid the chance of having continuous collisions. j The formula of collision rate fcurr in jth period is given by j fcurr ¼
Eðcollisionsj ½pÞ Eðdata sentj ½pÞ
ð2Þ
where the E(collisionj[p]) is the average number of collisions of the jth period for a user P and E(data_sentj[p]) is the average number of frames sent by the user. In order to create different contention windows for various priority classes, the AEDCF [1] uses the Multiplicator Factor (MF) to control its speed, based on moving averagj ing of fcurr and j MF½i ¼ minðð1 þ ði 2ÞÞ favg ; 0:8Þ
ð3Þ
with j j j1 favg ¼ ð1 aÞ fcurr þ a favg
ð4Þ
where i represents different priority class. The smaller value of i, the smaller the MF, i.e., the higher priority class, the smaller MF. After successful transmission the contention windows should not be greater than the original one. Then the contention windows for various priority class after successful delivery is given by CWnew ½i ¼ maxðCWmin ½i; CWold ½i MF½iÞ
ð5Þ
The above formula guarantees that the new contention window is greater than or equal to CWmin. On the other hand, the contention window after each collision is given by CWnew ½i ¼ minðCWmax ½i; CWold ½i PF½iÞ
ð6Þ
where high priority traffic flow has a smaller value of PF[i] for more chance of competition. 3. Improvement method of AEDCF: M-EDCF to reduce collisions In the IEEE 802.11e standard, if a channel is idle continuously for (AIFS+X) time slots, back_off timer can reduce X time slots, and system can start frame delivery after the back_off timer becomes zero. However, if the system detects a busy channel during the back_off time period, then it must stop the back_off procedure and set up the NAV (Virtual carrier sense). The problem is that a user only needs to wait for sufficient scattered idle time slots and then transmit after the back_off timer counts down to zero. Thus a low priority user may accumulate some idle time slots and may get the same privilege as a high priority user, which will result in higher collision rate. This is especially true if channel
844
W.-Y. Lin, J.-S. Wu / Computer Communications 30 (2007) 841–848
loading is very high, when an enormous amount of collisions cannot be avoided. Therefore, we hope to wait continuously during the whole idle period, so the system can start delivering after back_off timer count down to zero. Wherever a busy channel situation is detected in a backoff state, we must increase the contention window based on the average collision rate given by Eq. (4) and choose a new back_off time and start the backoff procedure. However, if we inflexibly increase the contention window, then the following situations may occur:
In order not to increase contention window rapidly, we limit the maximum value of the parameter ‘‘temp’’ to 2.0. In addition, we do not want the parameter ‘‘temp’’ to be less than 1, so the contention window decreases instead. The parameter b is called a scaling factor, which is used to control the increasing speed of contention window when channel loading is between low and high level. To describe our proposed scheme completely, a flow chart is presented in Fig. 3.
1. If the contention windows increase slightly, it still results in severe collisions when channel loading is high. 2. If the contention windows grows too much, this results in many idle periods when channel loading is low.
The simulation is based on the infrastructure mode of the 802.11e, where each working station generates three traffic flows (i.e., video, voice and background) delivered to the AP, and starts delivering at a random time. When the number of a working station increases, then channel loading increases. The MAC/PHY parameters are listed in Tables 2 and 3.
Therefore, we must adjust the contention window dynamically according to traffic load, which is based on the average collision rate. Thus as the average collision rate increases, that means traffic loading increases. The formula for the average collision rate follows Eq. (4). We present new contention windows as the following formula. where Temp ¼ minðb avg coll ; rate;2:0Þ New cw½pri ¼ old cw½pri maxðtemp;1:0Þ
ð7Þ ð8Þ
4. Simulation results
4.1. Case I First, we discuss the effect of parameter b toward the throughput. All the traffic, including voice, data and background are treated as CBR (constant bit rate), so that we can clearly observe the effect of the parameter b. The traffic parameters are listed in Table 4. Figs. 4 and 5 show system throughput vs b. As b increases, the total throughput increases because the average collision
Fig. 3. The Frame transmission flow chart.
W.-Y. Lin, J.-S. Wu / Computer Communications 30 (2007) 841–848 Table 2 802.11a PHY/MAC [10] parameters SIFS DIFS ACK size Data rate Slot time CCA time MAC header Modulation Preamble length Rx Tx turnaround taime PLCP header length
16 ls 34 ls 14 b 36 Mbps 9 ls 3 ls 28 b 16-QAM 20 ls 1 ls 4 ls
Table 3 802.11a PHY/MAC [10] parameters
Transport AC CWmin CWmax AIFSN
Voice
Video
Background
UDP 0 7 15 2
UDP 2 15 31 2
UDP 3 31 1023 3
rate is decreases when the traffic loading is high. We note that choosing b in the range of (5,14) achieves good performance. It seems that a good choice for b to be 9, no mater whether the traffic loading is high or low. In Fig. 5, the curve of ‘‘cw*1’’ stands for the scheme where the contention window size does not increase even where the collision rate increases, and ‘‘cw*2’’ stands for the scheme where the contention window size will double if the channel is sensed to be busy during backoff time. When the traffic load increases, the throughput curve of ‘‘cw*1’’ decreases because of the contention windows size is too small to avoid collisions. When the traffic load is low, the performance represented by the curve of ‘‘cw*2’’ is poorer than our proposed scheme because of long idle time. So the M-EDCF scheme has good performance for both traffic loading is high and low. Figs. 6–8 show that our proposed scheme M-EDCF can avoid the collisions better than the EDCF and the AEDCF
Table 4 The parameter of traffic in Case I Parameters
Voice
Video
Background
Packet size Packet interval Data rate Dgent
92 bytes 20 ms 34.4 kbps CBR
1464 bytes 10 ms 1171.2 kbps CBR
1500 bytes 10 ms 1200 kbps CBR Fig. 6. Total throughput in Case I.
Fig. 4. Total throughput vs b factor. Fig. 7. Channel utilization in Case I.
Fig. 5. Total throughput vs various schemes.
845
Fig. 8. Collision rate in Case I.
846
W.-Y. Lin, J.-S. Wu / Computer Communications 30 (2007) 841–848
schemes when traffic load is high. Thus it performs better in terms of throughput and utilization. In addition, voice and video data should be delay-sensitive. Figs. 9 and 10 show that the mean delay of ether voice or video data is less then 10 ms by using our proposed method. Compared with the EDCF and the AEDCF scheme, the M-EDCF performs much better. When the buffer is overflows, the frame is dropped out. In Case I, there are no drops of voice data, so we only present the dropping rates of background traffic and video traffic. Fig. 11 shows that the dropping rate in our method is less than both the EDCF and the AEDCF scheme.
4.2. Case II Traffic parameters in case II are showed in Table 5. Voice traffic is generated by the on/off model build in the NS2 module [23], while video is simulated by VBR (Variable Bit Rate) base on the trace produced by H.261 coding technology and QCIF resolution. We use CBR to simulate the Background traffic. Fig. 12.
Table 5 The parameters of traffic in Case II Voice
Agent Packet interval Packet size Data rate Burst_time Idle_time
Exponential 20 ms 160 bytes 64 kbps 400 ms 600 ms
Video
Agent Mean packet interval Mean data rate Mean packet size
VBR 26ms 200kbps 660bytes
Background
Agent Packet interval Packet size Data rate
CBR 20 ms 1600 bytes 640 kbps
Fig. 9. Mean delay of voice traffic in Case I.
Fig. 10. Mean delay of video traffic in Case I. Fig. 12. Drops per second of background traffic in Case I. 35 edcf
Drops per second
30
aedcf
25
proposed method
20 15 10 5 0 5
6
7
8
9 10 Station number
11
12
13
Fig. 11. Drops per second of video traffic in Case I.
14
Fig. 13. Total throughput in Case II.
W.-Y. Lin, J.-S. Wu / Computer Communications 30 (2007) 841–848
847
As shown in Figs. 13–15, the proposed M-EDCF scheme can provide better performance than the EDCF and the AEDCF in terms of throughput, channel utilization and collision rate. Figs. 16 and 17 show that the mean delay in the EDCF and the AEDCF scheme both increase quickly because of an increasing collision rate when the traffic load is high. But the curves of M-EDCF show that the mean delay is much smaller and increases gradually. (Figs. 18 and 19).
Fig. 14. Channel utilization in Case II.
Fig. 18. Mean delay of voice traffic in Case II.
Fig. 15. Collision rate in Case II.
Fig. 19. Mean delay of video traffic in Case II. Fig. 16. Mean delay of voice traffic in Case II.
Fig. 17. Mean delay of video traffic in Case II.
Fig. 20. Drops per second of video traffic in Case II.
848
W.-Y. Lin, J.-S. Wu / Computer Communications 30 (2007) 841–848
Fig. 21. Drops per second of background traffic in Case II.
When the buffer overflows in Case II, the dropping rate of voice data is also zero, so we present the results of background traffic and video traffic in Figs. 20 and 21, respectively. We can see that the dropping rate by our method is also lower than the EDCF and the AEDCF. Based on the simulation results showed in Case I and Case II, we can conclude that our proposed scheme, MEDCF, and the AEDCF scheme outperform the EDCF. Using the adaptive Back_off_timer, the M-EDCF provides much higher goodput than the ADCF scheme. Moreover, the M-EDCF scheme can improve the performance for both high priority and low priority traffic. 5. Conclusion Although there are numerous articles addressing the performance enhancement of the IEEE 802.11e EDCF, their proposals are either too complex or generate a great collisions under overload, which results in poor performance. In this paper, we provided a simple and effective method to reduce collisions, which can also distinguish between high priority and low priority traffic. Under the high traffic load, the throughput of high priority flow is protected. In addition, because the contention parameter is dynamically adjusted contention parameter, low priority traffic flow increases the contention window to lessen collisions under overloaded conditions. Compared with the EDCF and the AEDCF scheme, even low priority traffic flow can also obtain better performance by our proposed scheme. References [1] L. Romdhani, Ni. Qiang, T. Turletti, Adaptive EDCF: enhanced service differentiation for IEEE 802.11 wireless ad-hoc Networks, in: IEEE Wireless Communications and Networking, Conference, vol. 2, March (2003), pp. 16–20. [2] IEEE WG, Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications, IEEE 802.11 Standard (1999).
[3] IEEE 802.11e WG, Medium Access Control (MAC) Enhancements for Quality of Service, IEEE 802.11e Standard (2005). [6] G.Y. Liu, Wei Xin, An adaptive QoS guarantee scheme for multimedia wireless networks, in: Proceedings 2005 International Conference on Communications, Circuits and Systems, vol. 1, May (2005), pp. 353–357. [7] YuLong Fan, ChingYao Huang, YuRu Hong, Timer based scheduling control algorithm in WLAN for real-time service, IEEE International Symposium on Circuits and Systems 5 (2005) 4533– 4537. [8] Cheng Kuan, Ng ZhengYong, Providing QoS in congested IEEE 802.11 hot spots, 2005. DFMA ’05, in: First International Conference on Distributed Frameworks for Multimedia Applications, Feb. 2005, pp. 2–7. [9] D.-J. Deng, R.-S. Chang, A priority scheme for IEEE 802.11 DCF access method, IEICE Trans.Commun. E82-B (1) (1999) 96–102. [10] Y. Xiao, A simple and effective priority scheme for IEEE 802.11, IEEE Commun. Lett. 7 (2) (2003) 70–72. [11] A. Veres, A.T. Campbell, M. Barry, L.-H. Sun, Supporting differentiation in wireless packet networks using distributed control, IEEE J. Sel. Areas Commun. 19 (10) (2001) 2081–2093. [12] I. Aad, C. Castelluccia, Differentiation mechanisms for IEEE 802.11, in: IEEE Information Communications (INFOCOM), Anchorage, AK, 2001, pp. 209–218. [13] X. Pallot, L.E. Miller, Implementing message priority policies over an 802.11 based mobile ad hoc network, in: IEEE Military Communications Conf. (MILCOM), McLean, VA, 2001, pp. 860–864. [14] S. Mangold, S. Choi, P. May, O. Kein, G. Hiertz, L. Stibor, IEEE 802.11e Wireless LAN for quality of service, in: European Wireless, Florence, Italy, 2002, pp. 32–39. [15] Y. Xiao, IEEE 802.11e: A QoS provisioning at the MAC layer, IEEE Wireless Commun. 11 (3) (2004) 72–79. [16] Y. Xiao, H. Li, S. Choi, Protection and guarantee for voice and video traffic in IEEE 802.11eWireless LANs, in: Proceedings of the IEEE Information Communications (INFOCOM), Hong Kong, 2004, pp. 2153–2163. [22] I. Aad, C. Castelluccia, Remarks on Per-Flow Differentiation in IEEE 802.11, in: Proceedings of European Wireless, Feb 25th–28th, 2002. [23] Ni, Qiang-Romdhani, Lamia-Turletti, Thierry-Aad, Imad, QoS Issues and Enhancements for IEEE 802.11 Wireless LAN, INRIA Research Report No. 4612, Nov. 2002.
Wen Yen Lin received the MS degrees in the Dept. of Electrical Engineering from National Central University in 199, and is currently a Ph.D. candidate in the Dept. of Electrical Engineering National Central University, Taiwan, R.O.C . He is also now a lecturer in the Dept. of Electronic Engineering at Vanung University, Taiwan, Republic of China. His research interests include mobile computing, ad hoc wireless networks and multimedia communications.
Jung-Shyr Wu received his Ph.D. in electrical engineering from the University of Calgary Canada in 1989. He is a full Professor in the Graduate institute of Communication Engineering at National Central University, Taiwan, Republic of China. His research interests include computer networks, wireless networks, mobile communication and queueing theory.
Modified EDCF to improve the performance of IEEE 802.11e WLAN Wen-Yen Lin
a,b
, Jung-Shyr Wu
c,*
a
c
Department of Electrical Engineering, National Central University, Taiwan, ROC b Department of Electronic Engineering, Vanung University, Taiwan, ROC Department of Communication Engineering, National Central University, Taiwan, ROC
Received 22 December 2005; received in revised form 11 October 2006; accepted 19 October 2006 Available online 21 November 2006
Abstract In this paper, we propose a modified EDCF scheme, M-EDCF, to improve the Quality of Service (QoS) of the IEEE 802.11e wireless network. The IEEE 802.11e standard is presented to support QoS at medium access control level using a priority scheme by differentiating the inter-frame space and the initial window size. In addition to providing relative priorities by adjusting the size of the Contention Window (CW) of each traffic class, our proposed scheme, M-EDCF, also consider the effect of a back_off_timer to avoid unnecessary collisions. Our study shows that in either in heavy or light traffic load our proposed scheme can provide better quality for both high priority and low priority packets than either the AEDCF [L. Romdhani, Ni. Qiang, T. Turletti, Adaptive EDCF: enhanced service differentiation for IEEE 802.11 wireless ad-hoc networks, IEEE Wireless Communications and Networking, Conference. vol. 2, March (2003) pp. 16–20] or the original EDCF. Ó 2006 Elsevier B.V. All rights reserved. Keywords: 802.11e; Enhance distributed coordination function (EDCF); Quality of service (QoS); Wireless LAN; 802.11
1. Introduction The IEEE 802.11 enables fast installation, with minimum management and maintenance costs, and is a very robust protocol for the best-effort service in the wireless medium. However, it is unsuitable for multimedia applications with Quality of Service (QoS) requirements. The IEEE 802.11 medium access control (MAC) [2] employs a mandatory contention-based channel access function called distributed coordination function (DCF), and an optional centrally controlled channel access function called point coordination functions (PCF). The DCF adopts a carrier sense multiple access with collision avoidance (CSMA/ CA) with binary exponential backoff. However it does not support QoS, and cannot support any priority scheme even through the PCF can support very limited QoS, as an
*
Corresponding author. Tel.: +886 913976255; fax: +886 34514141. E-mail addresses: [email protected], [email protected] (J.-S. Wu).
0140-3664/$ - see front matter Ó 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.comcom.2006.10.013
optional function that has not been implemented in the most of current products. Since its introduction, various extensions and modifications have been studied to address this current need and the IEEE 802.11 Task Group E [3] is responsible for developing a QoS-aware MAC protocol that considers several service differentiation mechanisms. However, the performance of service differentiation has only been evaluated for a supplement to the original IEEE 802.11 MAC. The IEEE 802.11e MAC will support multimedia applications such as voice and video over the IEEE 802.11 WLAN. The IEEE 802.11e MAC also employs a contentionbased channel access function called the enhanced distributed coordination function (EDCF), and a centrally controlled channel access function called the Hybrid Coordination Function (HCF). The EDCF provides a priority scheme by differentiating the inter-frame space and the initial window size. Such a priority scheme may provide prioritized but not guaranteed QoS due to it’s contention-based nature. Prioritized QoS will be useful for those
842
W.-Y. Lin, J.-S. Wu / Computer Communications 30 (2007) 841–848
multimedia applications that do not require rigid QoS. One advantage of prioritized QoS is that it is simple to implement since it looks like the DiffServ model. In this article, we focus on the EDCF, but not the HCF. Quality of Service (QoS) support is critical to multimedia applications. Here, time-bounded services such as audio and video conference typically require some specified bandwidth, delay and jitter guarantee, but can tolerate some losses. However, in DCF all the stations in a Basic Service Set or all the flows from the same station compete for resources and channels with the same priority. There is no differentiation mechanism to guarantee packet delay and jitter to stations or flows to supporting time-bounded multimedia services. Several priority studies have been reported in the literature for DCF and EDCF, and priority schemes can be classified into three kinds: Backoff based priority schemes [9–11], IFS-based priority schemes [12,13], and hybrid priority schemes [1,6–8,14–16]. The hybrid schemes, such as the EDCF [3], adopt both a backoff based scheme and an IFS-based scheme, and most recent reports focus on it [1,6–8]. Although those mechanisms improve the quality of service of real-time traffic, the performance levels obtained are not optimal since EDCF parameters cannot be adapted to the network conditions. When the collision rate increases very fast, the contentions to access the shared medium are very high, which significantly affects the throughput. This motivates us to propose a scheme that adapts the CW parameter and back_off_timer according to the network conditions. Our proposed scheme is called the M-EDCF, and is a hybrid scheme based on Ref. [1] to improve the performance of throughput, mean delay and dropping rate in the buffer. In this paper, we introduce the 802.11e EDCF priority scheme and AEDCF in Section 2. In Section 3, we describe our proposed scheme, called the modified AEDCF priority (MA_EDCF) scheme. Simulation results are studied in Section 4. Section 5 concludes the paper by summarizing results and outlining future research.
2. IEEE 802.11e and adaptive EDCF The 802.11 legacy MAC does not support the concept of differentiating frames with different priorities. Basically, the DCF is supposed to provide channel access with equal probabilities to all stations contending for the channel access in a distributed manner. However, equal access probabilities are not desirable among stations with different priority frames. The emerging EDCF is designed to provide differentiated and distributed channel access for frames with eight different priorities (from 0 to 7) by enhancing the DCF, as shown in Fig. 1. In contrast to original DCF, the EDCF is not a separate coordination function. Rather, it is a part of a single coordination function, called the Hybrid Coordination Function (HCF), of the 802.11e MAC. The HCF combines the aspects of both DCF and PCF. All the detailed aspects of the HCF are beyond the scope of this paper since we focus on the HCF contention-based channel access, i.e., EDCF. Each frame from the higher layer arrives at the MAC along with a specific priority value. Then each QoS data frame carries its priority value in the MAC frame header. An 802.11e STA implements four access categories (ACs), where an AC is an enhanced variant of the DCF 0. Each frame arriving at the MAC with a priority is mapped into an AC as shown in Table 1. Table 1 Category mappings of priority to access Priority
Access category (AC)
Designation (informative)
1 2 0 3 4 5 6 7
0 0 0 1 2 2 3 3
Best effort Best effort Best effort Video probe Video Video Video Video
Fig. 1. Basic DCF vs queue-based EDCF.
W.-Y. Lin, J.-S. Wu / Computer Communications 30 (2007) 841–848
2.1. HCCA (polling-based-HCF channel access scheme) The HCCA is similar to the channel access scheme of PCF in that both are served by AP. When using HCCA, the AP can obtain the control of wireless channel after a time interval of PIFS, and start up the working station in the polling area, where information needs to be delivered. Before this, the AP must obtain traffic flow information for each working station through controlling frames so that AP can use this information to arrange the wireless sources. As in PCF, HCCA can also establish the Superframe, which is composed of CP-contention phase and CFP-contention free phase. At the same time, the AP is allowed to initiate delivery at any time when information needs to be delivered. 2.2. EDCA (contention-based-EDCF) Basically, an AC uses AIFS[AC], CWmin[AC], and CWmax[AC] instead of DIFS, CWmin, and CWmax, of the DCF, respectively, for the contention process to transmit a frame belonging to access category AC, as shown in Fig. 2.The AIFS[AC] is determined by AIFS½AC ¼ SIFS þ AIFS½AC SlotTime
ð1Þ
where AIFS[AC] is an integer greater than zero. Moreover, the backoff counter is selected from [1, 1+CW[AC]]. 2.3. Related improved schemes for the IEEE 802.11 To date, there are many studies on 802.11 wireless networks supplied with instant information as quality guarantee. They are classified into two groups, one is station-based improvement [11,12], and the other one is queue-based improvement [22]. The former indicates represents that each working station has its own special parameter, and the latter represents that each working station has many queues, which serve as simulated working stations, and each queue has its own parameter. 2.4. AEDCF (adaptive EDCF) illustration After each successful delivery, the EDCF will reset its contention window to CWmin, regardless of present
Fig. 2. The EDCF channel access.
843
network conditions. However, after a collision occurs, the chance of having the second collision will be higher within a short time, so the method mentioned in [1] is based on the network condition, and gradually lessens the contention window instead of resetting the windows value to CWmin directly to avoid the chance of having continuous collisions. j The formula of collision rate fcurr in jth period is given by j fcurr ¼
Eðcollisionsj ½pÞ Eðdata sentj ½pÞ
ð2Þ
where the E(collisionj[p]) is the average number of collisions of the jth period for a user P and E(data_sentj[p]) is the average number of frames sent by the user. In order to create different contention windows for various priority classes, the AEDCF [1] uses the Multiplicator Factor (MF) to control its speed, based on moving averagj ing of fcurr and j MF½i ¼ minðð1 þ ði 2ÞÞ favg ; 0:8Þ
ð3Þ
with j j j1 favg ¼ ð1 aÞ fcurr þ a favg
ð4Þ
where i represents different priority class. The smaller value of i, the smaller the MF, i.e., the higher priority class, the smaller MF. After successful transmission the contention windows should not be greater than the original one. Then the contention windows for various priority class after successful delivery is given by CWnew ½i ¼ maxðCWmin ½i; CWold ½i MF½iÞ
ð5Þ
The above formula guarantees that the new contention window is greater than or equal to CWmin. On the other hand, the contention window after each collision is given by CWnew ½i ¼ minðCWmax ½i; CWold ½i PF½iÞ
ð6Þ
where high priority traffic flow has a smaller value of PF[i] for more chance of competition. 3. Improvement method of AEDCF: M-EDCF to reduce collisions In the IEEE 802.11e standard, if a channel is idle continuously for (AIFS+X) time slots, back_off timer can reduce X time slots, and system can start frame delivery after the back_off timer becomes zero. However, if the system detects a busy channel during the back_off time period, then it must stop the back_off procedure and set up the NAV (Virtual carrier sense). The problem is that a user only needs to wait for sufficient scattered idle time slots and then transmit after the back_off timer counts down to zero. Thus a low priority user may accumulate some idle time slots and may get the same privilege as a high priority user, which will result in higher collision rate. This is especially true if channel
844
W.-Y. Lin, J.-S. Wu / Computer Communications 30 (2007) 841–848
loading is very high, when an enormous amount of collisions cannot be avoided. Therefore, we hope to wait continuously during the whole idle period, so the system can start delivering after back_off timer count down to zero. Wherever a busy channel situation is detected in a backoff state, we must increase the contention window based on the average collision rate given by Eq. (4) and choose a new back_off time and start the backoff procedure. However, if we inflexibly increase the contention window, then the following situations may occur:
In order not to increase contention window rapidly, we limit the maximum value of the parameter ‘‘temp’’ to 2.0. In addition, we do not want the parameter ‘‘temp’’ to be less than 1, so the contention window decreases instead. The parameter b is called a scaling factor, which is used to control the increasing speed of contention window when channel loading is between low and high level. To describe our proposed scheme completely, a flow chart is presented in Fig. 3.
1. If the contention windows increase slightly, it still results in severe collisions when channel loading is high. 2. If the contention windows grows too much, this results in many idle periods when channel loading is low.
The simulation is based on the infrastructure mode of the 802.11e, where each working station generates three traffic flows (i.e., video, voice and background) delivered to the AP, and starts delivering at a random time. When the number of a working station increases, then channel loading increases. The MAC/PHY parameters are listed in Tables 2 and 3.
Therefore, we must adjust the contention window dynamically according to traffic load, which is based on the average collision rate. Thus as the average collision rate increases, that means traffic loading increases. The formula for the average collision rate follows Eq. (4). We present new contention windows as the following formula. where Temp ¼ minðb avg coll ; rate;2:0Þ New cw½pri ¼ old cw½pri maxðtemp;1:0Þ
ð7Þ ð8Þ
4. Simulation results
4.1. Case I First, we discuss the effect of parameter b toward the throughput. All the traffic, including voice, data and background are treated as CBR (constant bit rate), so that we can clearly observe the effect of the parameter b. The traffic parameters are listed in Table 4. Figs. 4 and 5 show system throughput vs b. As b increases, the total throughput increases because the average collision
Fig. 3. The Frame transmission flow chart.
W.-Y. Lin, J.-S. Wu / Computer Communications 30 (2007) 841–848 Table 2 802.11a PHY/MAC [10] parameters SIFS DIFS ACK size Data rate Slot time CCA time MAC header Modulation Preamble length Rx Tx turnaround taime PLCP header length
16 ls 34 ls 14 b 36 Mbps 9 ls 3 ls 28 b 16-QAM 20 ls 1 ls 4 ls
Table 3 802.11a PHY/MAC [10] parameters
Transport AC CWmin CWmax AIFSN
Voice
Video
Background
UDP 0 7 15 2
UDP 2 15 31 2
UDP 3 31 1023 3
rate is decreases when the traffic loading is high. We note that choosing b in the range of (5,14) achieves good performance. It seems that a good choice for b to be 9, no mater whether the traffic loading is high or low. In Fig. 5, the curve of ‘‘cw*1’’ stands for the scheme where the contention window size does not increase even where the collision rate increases, and ‘‘cw*2’’ stands for the scheme where the contention window size will double if the channel is sensed to be busy during backoff time. When the traffic load increases, the throughput curve of ‘‘cw*1’’ decreases because of the contention windows size is too small to avoid collisions. When the traffic load is low, the performance represented by the curve of ‘‘cw*2’’ is poorer than our proposed scheme because of long idle time. So the M-EDCF scheme has good performance for both traffic loading is high and low. Figs. 6–8 show that our proposed scheme M-EDCF can avoid the collisions better than the EDCF and the AEDCF
Table 4 The parameter of traffic in Case I Parameters
Voice
Video
Background
Packet size Packet interval Data rate Dgent
92 bytes 20 ms 34.4 kbps CBR
1464 bytes 10 ms 1171.2 kbps CBR
1500 bytes 10 ms 1200 kbps CBR Fig. 6. Total throughput in Case I.
Fig. 4. Total throughput vs b factor. Fig. 7. Channel utilization in Case I.
Fig. 5. Total throughput vs various schemes.
845
Fig. 8. Collision rate in Case I.
846
W.-Y. Lin, J.-S. Wu / Computer Communications 30 (2007) 841–848
schemes when traffic load is high. Thus it performs better in terms of throughput and utilization. In addition, voice and video data should be delay-sensitive. Figs. 9 and 10 show that the mean delay of ether voice or video data is less then 10 ms by using our proposed method. Compared with the EDCF and the AEDCF scheme, the M-EDCF performs much better. When the buffer is overflows, the frame is dropped out. In Case I, there are no drops of voice data, so we only present the dropping rates of background traffic and video traffic. Fig. 11 shows that the dropping rate in our method is less than both the EDCF and the AEDCF scheme.
4.2. Case II Traffic parameters in case II are showed in Table 5. Voice traffic is generated by the on/off model build in the NS2 module [23], while video is simulated by VBR (Variable Bit Rate) base on the trace produced by H.261 coding technology and QCIF resolution. We use CBR to simulate the Background traffic. Fig. 12.
Table 5 The parameters of traffic in Case II Voice
Agent Packet interval Packet size Data rate Burst_time Idle_time
Exponential 20 ms 160 bytes 64 kbps 400 ms 600 ms
Video
Agent Mean packet interval Mean data rate Mean packet size
VBR 26ms 200kbps 660bytes
Background
Agent Packet interval Packet size Data rate
CBR 20 ms 1600 bytes 640 kbps
Fig. 9. Mean delay of voice traffic in Case I.
Fig. 10. Mean delay of video traffic in Case I. Fig. 12. Drops per second of background traffic in Case I. 35 edcf
Drops per second
30
aedcf
25
proposed method
20 15 10 5 0 5
6
7
8
9 10 Station number
11
12
13
Fig. 11. Drops per second of video traffic in Case I.
14
Fig. 13. Total throughput in Case II.
W.-Y. Lin, J.-S. Wu / Computer Communications 30 (2007) 841–848
847
As shown in Figs. 13–15, the proposed M-EDCF scheme can provide better performance than the EDCF and the AEDCF in terms of throughput, channel utilization and collision rate. Figs. 16 and 17 show that the mean delay in the EDCF and the AEDCF scheme both increase quickly because of an increasing collision rate when the traffic load is high. But the curves of M-EDCF show that the mean delay is much smaller and increases gradually. (Figs. 18 and 19).
Fig. 14. Channel utilization in Case II.
Fig. 18. Mean delay of voice traffic in Case II.
Fig. 15. Collision rate in Case II.
Fig. 19. Mean delay of video traffic in Case II. Fig. 16. Mean delay of voice traffic in Case II.
Fig. 17. Mean delay of video traffic in Case II.
Fig. 20. Drops per second of video traffic in Case II.
848
W.-Y. Lin, J.-S. Wu / Computer Communications 30 (2007) 841–848
Fig. 21. Drops per second of background traffic in Case II.
When the buffer overflows in Case II, the dropping rate of voice data is also zero, so we present the results of background traffic and video traffic in Figs. 20 and 21, respectively. We can see that the dropping rate by our method is also lower than the EDCF and the AEDCF. Based on the simulation results showed in Case I and Case II, we can conclude that our proposed scheme, MEDCF, and the AEDCF scheme outperform the EDCF. Using the adaptive Back_off_timer, the M-EDCF provides much higher goodput than the ADCF scheme. Moreover, the M-EDCF scheme can improve the performance for both high priority and low priority traffic. 5. Conclusion Although there are numerous articles addressing the performance enhancement of the IEEE 802.11e EDCF, their proposals are either too complex or generate a great collisions under overload, which results in poor performance. In this paper, we provided a simple and effective method to reduce collisions, which can also distinguish between high priority and low priority traffic. Under the high traffic load, the throughput of high priority flow is protected. In addition, because the contention parameter is dynamically adjusted contention parameter, low priority traffic flow increases the contention window to lessen collisions under overloaded conditions. Compared with the EDCF and the AEDCF scheme, even low priority traffic flow can also obtain better performance by our proposed scheme. References [1] L. Romdhani, Ni. Qiang, T. Turletti, Adaptive EDCF: enhanced service differentiation for IEEE 802.11 wireless ad-hoc Networks, in: IEEE Wireless Communications and Networking, Conference, vol. 2, March (2003), pp. 16–20. [2] IEEE WG, Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications, IEEE 802.11 Standard (1999).
[3] IEEE 802.11e WG, Medium Access Control (MAC) Enhancements for Quality of Service, IEEE 802.11e Standard (2005). [6] G.Y. Liu, Wei Xin, An adaptive QoS guarantee scheme for multimedia wireless networks, in: Proceedings 2005 International Conference on Communications, Circuits and Systems, vol. 1, May (2005), pp. 353–357. [7] YuLong Fan, ChingYao Huang, YuRu Hong, Timer based scheduling control algorithm in WLAN for real-time service, IEEE International Symposium on Circuits and Systems 5 (2005) 4533– 4537. [8] Cheng Kuan, Ng ZhengYong, Providing QoS in congested IEEE 802.11 hot spots, 2005. DFMA ’05, in: First International Conference on Distributed Frameworks for Multimedia Applications, Feb. 2005, pp. 2–7. [9] D.-J. Deng, R.-S. Chang, A priority scheme for IEEE 802.11 DCF access method, IEICE Trans.Commun. E82-B (1) (1999) 96–102. [10] Y. Xiao, A simple and effective priority scheme for IEEE 802.11, IEEE Commun. Lett. 7 (2) (2003) 70–72. [11] A. Veres, A.T. Campbell, M. Barry, L.-H. Sun, Supporting differentiation in wireless packet networks using distributed control, IEEE J. Sel. Areas Commun. 19 (10) (2001) 2081–2093. [12] I. Aad, C. Castelluccia, Differentiation mechanisms for IEEE 802.11, in: IEEE Information Communications (INFOCOM), Anchorage, AK, 2001, pp. 209–218. [13] X. Pallot, L.E. Miller, Implementing message priority policies over an 802.11 based mobile ad hoc network, in: IEEE Military Communications Conf. (MILCOM), McLean, VA, 2001, pp. 860–864. [14] S. Mangold, S. Choi, P. May, O. Kein, G. Hiertz, L. Stibor, IEEE 802.11e Wireless LAN for quality of service, in: European Wireless, Florence, Italy, 2002, pp. 32–39. [15] Y. Xiao, IEEE 802.11e: A QoS provisioning at the MAC layer, IEEE Wireless Commun. 11 (3) (2004) 72–79. [16] Y. Xiao, H. Li, S. Choi, Protection and guarantee for voice and video traffic in IEEE 802.11eWireless LANs, in: Proceedings of the IEEE Information Communications (INFOCOM), Hong Kong, 2004, pp. 2153–2163. [22] I. Aad, C. Castelluccia, Remarks on Per-Flow Differentiation in IEEE 802.11, in: Proceedings of European Wireless, Feb 25th–28th, 2002. [23] Ni, Qiang-Romdhani, Lamia-Turletti, Thierry-Aad, Imad, QoS Issues and Enhancements for IEEE 802.11 Wireless LAN, INRIA Research Report No. 4612, Nov. 2002.
Wen Yen Lin received the MS degrees in the Dept. of Electrical Engineering from National Central University in 199, and is currently a Ph.D. candidate in the Dept. of Electrical Engineering National Central University, Taiwan, R.O.C . He is also now a lecturer in the Dept. of Electronic Engineering at Vanung University, Taiwan, Republic of China. His research interests include mobile computing, ad hoc wireless networks and multimedia communications.
Jung-Shyr Wu received his Ph.D. in electrical engineering from the University of Calgary Canada in 1989. He is a full Professor in the Graduate institute of Communication Engineering at National Central University, Taiwan, Republic of China. His research interests include computer networks, wireless networks, mobile communication and queueing theory.