Project Report: QoS Enhancement for Real-Time Traffic in IEEE802.11 WLAN Abstract A key issue in IEEE802.11 WLAN MAC is how to provide QoS support, especially for time-bounded traffic. Although much work has been done and pretty good performance has been achieved, QoS of real-time traffic is still not guaranteed. In this paper, we review the development of contention-based IEEE802.11 MAC schemes and compare their pros and cons. In order to provide a better support for real-time QoS, we propose a new scheme that differentiates real-time traffic from other prioritized traffics and eliminates the possible collisions between them. Simulation results show that our scheme can reduce the access delay of real-time traffic to a maximum of about 60% with little cost of network throughput. 1 Introduction The increasingly wide use of wireless local area networks has aroused much research work on the Quality-of-Service issues in the existent IEEE802.11 MAC protocol. Since the demand of multimedia transmission over WLAN is continuously growing, the QoS of real-time traffics becomes a critical issue. Good overviews of the issue can be found in [2], [3]. In the original IEEE802.11 standard, services are not differentiated and hence no support for real-time traffic is provided. The newly issued IEEE802.11e makes considerable improvement on the differentiation of services, but the performance is still not satisfactory enough in some aspects. Recent works have been focused on how to increase throughput, decrease latency and achieve better fairness. There have been a number of schemes proposed for that purpose. ([6] and [7] did simulation and evaluation on some of the early schemes.) However, the QoS of real-time traffic is never guaranteed. In this paper, we propose an approach based on previous works, which aims at enhancing the QoS of real-time traffic with no access point present. It treats the real-time traffics separately from non real-time ones so that no contention between them will occur. The rest of the report is organized as follows: In part 2, the gradual modifications on the original IEEE802.11 MAC are described to give an overview of the related work. Part 3 describes the proposed scheme and the expected improvement it may bring. The simulation results and analysis are presented in part 4 and we finally conclude in part 5. 2 Related Works - 1 -
2.1 DCF IEEE802.11 defines two mechanisms for medium access control: DCF (Distributed Coordinator Function) is used during contention period in both Ad-hoc and architecture network configurations, whereas PCF (Point Coordinator Function) is the access method during contention free period with the presence of a point coordinator (access point). In other words, PCF can be used only in architecture network. In this report, we focus on the improvement on DCF, which has a broader use and fewer requirements on the configuration of the network. DCF is based on CSMA/CA (Carrier Sense Multiple Access/Collision Avoidance). Basically, when the channel is sensed idle for more than a period of time called DIFS (Distributed Inter-Frame Space), the station can send its packet immediately. If it s sensed busy, the stations with packets to send wait until the channel becomes idle again and wait another DIFS period plus a backoff time. After the backoff timer times out, the station can finally transmit its packet. Suppose the channel becomes busy when a station is in its backoff time, the station pauses the backoff timer instead of stopping it, and will resume it once the channel comes back to idle state. Thus, a station which has been waiting a long time doesn t have to do that all over again. For the consideration of low collision rate and fairness, the backoff time is chosen randomly from the range of a so-called Contention Window (CW). The CW varies depending on whether the station transmitted successfully or not. After a successful transmission, the CW is reset to a minimum value CW min. In contrast, if a collision occurs, the CW is doubled until it reaches the maximum value CW max. As a summary, DCF is good for asynchronous data transmission, but it suffers significant performance degradation at high load conditions [7], because of the higher collision rate and wasted time on negotiations. Plus, DCF does not differentiate services, thus it s not suitable for time-bounded traffics. 2.2 EDCF To handle service differentiation, IEEE 802.11 was extended to 802.11b, in which Enhanced DCF (EDCF) is deployed as the contention based media access mechanism [8]. Its main modification on DCF is that eight levels of user priorities can be applied to stations. A station with higher priority is assigned shorter CW min and CW max, so that in most cases higher priority flows will have more chances to transmit before lower priority ones. Moreover, different IFS are introduced to different priority levels, which means higher priority flows have longer IFS, while the lower priority ones have shorter IFS. The IFS here is called Arbitrary IFS (AIFS). EDCF achieves a good prioritization. However it still has the same problem as DCF that it performs poorly when the traffic load is high due to frequent collisions and wasted idle time. Furthermore, EDCF suffers from low priority traffic starvation especially at high load conditions, which impairs its fairness. 2.3 AEDCF - 2 -
Service differentiation used in EDCF provides better services to high priority class while offering a minimum service for low priority traffic. Although this mechanism improves the Quality-of-Service of real-time traffic, EDCF parameters cannot be adapted to the network conditions, such as the collision rate and the network load. Adaptive EDCF (AEDCF) was proposed [9] to adapt the CW parameter according to the network conditions, so that better support for QoS is provided. The idea is as follows. After a successful transmission, the CW is updated slowly instead of being reset to CW min, which is for the purpose of avoiding busty collisions. Similarly, after a collision, the new CW is not doubled but increased with a persistence factor, causing that the CW of high priority traffic increases slower than that of low priority traffic. The gradual update of the CW takes into account the average collision rate at each station, which is computed periodically. The factor of CW update is calculated in such a way that flows with high collision rate will have a better chance to transmit next time. AEDCF successfully decreases the collision rate between stations with the same priority, and decreases the access latency as well. The problem of AEDCF is that the performance of background low priority flows degrades at high load, because the background traffic will have much larger average CW size than high priority traffics, thus increasing waiting time and impairing channel utilization. 2.4 AFEDCF From analysis of the above schemes, it can be seen that the main performance impairment of distributed contention-based approaches comes from packet collisions and wasted idle slots due to backoff in each contention cycle. The ideal case is reached when a successful packet transmission is followed by another successful packet transmission without any collisions or idle time loss. Aiming at decreasing collision and idle time, AFEDCF [10] deploys the mechanism in which the CW is increased not only when there s a collision but also when the medium is sensed busy during deferring periods. The backoff timer is decreased when the medium is sensed idle in two different stages: linear decrease and fast decrease. In the linear decrease, the backoff timer goes down one by one, while in the fast decrease, it decreases exponentially. The boundary between those two stages is the varying backoff threshold. Considering the traffic load, it should increase during low contention periods but decrease during high contention periods. From another perspective, when a collision occurs or the station is in the deferring period waiting for the channel to be idle, it doubles the CW, randomly chooses a new backoff time and reduces the backoff threshold to make the fast decrease phase shorter. After a successful transmission, the station resets the CW to CW min, randomly chooses the backoff time and increases the backoff threshold to make the fast decrease stage longer. AFEDCF doesn t use the adaptive CW update as AEDCF does. AFEDCF shows good performance in almost all respects. It achieves high throughput and fairness even at high load conditions. The fairness is due to the fact that if the traffic load is high, the CW of all stations reach rapidly their maximum values, thus they will be transmitting almost all the time at the same contention window. The deployment of varying backoff threshold contributes to less wasted idle time and adaptation to collision rate. - 3 -
3 Proposed Scheme Despite the pretty good performance of AFEDCF, it doesn t provide guarantees for real-time traffics. Although real-time traffic has a higher priority, it still takes the risk of losing the contention when competing with other traffics. For instance in the AFEDCF scheme, the real-time traffic (R) is assigned the highest priority with the smallest CW, say, CW min = 3, CW max = 15. The second highest priority (S) has the settings of CW min = 15, CW max = 31. The two priority levels have the same AIFS value, say 1. It is likely that at a certain point of time, the backoff time of R randomly chosen from [1, CW R + 1] is larger than that of S randomly chosen from [1, CW S + 1]. In this case, if R and S both try to transmit after the channel turns from busy to idle, R will lose. This is the problem we ve been trying to solve. In this section, we propose a scheme to further improve QoS of real-time traffic on the basis of AFEDCF. From the analysis of previous schemes, the observation is that simple prioritization is not enough. Real-time traffic might lose the competition because it must do the same backoff procedure which is undeterministic. So in our scheme, differentiation is extended. Real-time traffic is differentiated from other traffics by having no backoff period and the shortest AIFSN (number of time slots in AIFS). Thus, real-time traffic is guaranteed to transmit before other traffics if they re ready to transmit at the same time. Only collision between real-time traffics is possible. To avoid collisions between real-time traffics, they are assigned different values of AIFS, which take into account the time they have been waiting with the consideration of fairness included. In other words, the flow that has waiting a longer time will get a better chance to transmit compared to another real-time flow. The waiting time here is defined as the time since its first attempt to transmit. The probability of that two flows try to transmit for the first time at the same time and neither succeeds is quite small. The AIFS time of real-time traffic is bounded no more than AIFSN real timeslots. We compute the value of AIFS time as follows. Suppose T max is the maximum tolerable delay for time-bounded traffic, T wait is the time the real-time traffic has been waiting. If T wait >= T max, AIFS equals the bottom boundary, which is the shortest IFS that a flow can ever have. Otherwise, AIFS is a linear function of T wait. Expressed in equations, the computation is as follows: T AIFS = AIFSN AIFSN α, T < T wait real real wait Tmax AIFS = (1 α) AIFSN, T T Here, α is a factor for adjusting the range of AIFS time. We can see that with the inclusion of T wait, fairness between real-time traffics is provided. real wait max max Furthermore, an additional modification might be that non-real-time traffics can be interrupted by real-time traffic, with the contention window of the interrupted traffic reset to the minimum value. The backoff time used to be chosen from [1, CW+1], but is now chosen from [1*S pack /S max, (CW+1) *S pack /S max ], taking into consideration the packet size S pack. So the interrupted packet will have a big chance to resume next time. - 4 -
However, this modification was not implemented in our simulation, but it might be something worthy to try. 4 Simulations Our simulation is based on the previous work done by Qiang Ni et al. at INRIA Sophia Antipolis, France. We implemented our scheme in ns-2 and experimented it in an ad-hoc network configuration with different channel loads represented by the number of nodes in the network (4, 6 16). There s always a node serving as a common receiver, to which all other nodes send three different flows. Since we wanted to compare the performance of our scheme to that of AFEDCF, the three flows for AFEDCF simulation are background, medium and high, while the ones for our scheme are background, high and real-time. Real-time is the counterpart of high, because they are both the highest priority in their corresponding schemes. We use the same parameters for them except priority levels so that they can be compared with each other in terms of performance improvement or degradation. The network topology of the simulation can be found in figure 1. And the parameters used are listed in table 1. We focus on the comparison between our scheme and AFEDCF in the simulation. Figure 1. Simulation topology. Real-Time High (ours) (AFEDCF) Medium Background Priority level RT 3 2 0 CW min 0 7 15 31 CW max 0 15 31 1023 AIFSN 1(max) 1 1 2 Packet Size 160B 160B 1280B 1500B Packet Interval 20ms 20ms 20ms 12.5ms Table 1. Parameters used in the simulation. Figure 2 shows the total goodput for both AFEDCF and our scheme (which is called Real-time AFEDCF in the figure). Total goodput is the sum of the goodputs of the stations sending data. That s why it increases with the number of stations. Our scheme impairs a little bit the total goodput when the traffic load is high, because real-time traffic with smaller packet size transmits more, which more or less depends on the - 5 -
parameters used in the simulation. It can be seen that the impairment is tolerable. Figure 2. Comparison of total goodputs. Figure 3. Comparison of goodputs for three flows. - 6 -
Figure 4. Comparison of latencies for Real-Time/high flows. Figure 5. Comparison of latencies for medium flows. - 7 -
Figure 6. Comparison of latencies for background flows. The goodputs of the individual flows in the two schemes are plotted in Figure 3. The dotted lines depict the goodputs of the flows in our scheme. Only drops in background flows are observable. That is one of the costs of our improvement on the latency of real-time traffic. Other costs are slightly higher latency of medium flows and background flows, which can be seen from Figure 5 and Figure 6. This is self-explanatory. The lower priority flows lose the small chance they used to have to compete with real-time traffic. Sacrificing a little in other traffics, real-time traffic gains a significant improvement on access delay, which decreases about 60% in our scheme as shown in Figure 4. This is because we ve cancelled the backoff time for real-time traffic but let it backoff in the AIFS period, thus saving the wasted idle time. It s also due to the elimination of collisions between real-time traffic and non real-time traffic. Compared with the performance loss in non real-time traffic, the gain is more notable. So the sacrifice is worthwhile. 5 Conclusions In this report, we described an improvement on the existent IEEE802.11 wireless LAN MAC schemes, which differentiates real-time traffic from non real-time ones and eliminates the chance of collision between them. In the existent schemes, it s still possible that highest priority traffic collides with other priority traffic or even loses the contention. In contrast, our scheme guarantees that real-time traffic - 8 -
doesn t have to compete with non-priority traffic and always transmit first when the medium is idle. Simulation is done with different traffic loads, and the result shows that our scheme reduces the access latency of real-time traffic without considerable impairment of the performances of other traffics. References [1] B. Crown. et al., IEEE 802.11 wireless local area networks, IEEE Communications Magazine, Sep. 1997 [2] S. Sharma, Analysis of 802.11b MAC: A QoS, fairness, and performance perspective. www.ecsl.cs.sunysb.edu/tr/wlanrpe.pdf [3] Anders Lindgren, Andreas Almquist, and Olov Schelén, Quality of Service Schemes for IEEE 802.11, Mobile Networks and Applications, Volume 8, Issue 3 (June 2003) [4] I. Aad and C. Castelluccia, Differentiation mechanisms for IEEE 802.11, in Proceedings of IEEE INFOCOM 2001, Apr. 2001 [5] Priyank Garg, Rushabh Doshi, Russell Greene, Mary Baker, Majid Malek and Maggie Cheng, Using IEEE 802.11e MAC for QoS over wireless, IPCCC '03 [6] Anders Lindgren, Andreas Almquist, Olov Schelén, Quality of Service Schemes for IEEE 802.11: A Simulation Study, Proceedings of the 9th International Workshop on Quality of Service, p.281-287, June 06-08, 2001 [7] Anders Lindgren, Andreas Almquist, Olov Schelén, Quality of service schemes for IEEE 802.11 wireless LANs: an evaluation, Mobile Networks and Applications, v.8 n.3, p.223-235, June 2003 [8] Daqing Gu, Jinyun Zhang, QoS enhancement in IEEE802.11 WLAN, IEEE Communications, Jun. 2003. [9] Lamia Romdhani, Qiang Ni, and Thierry Turletti, "Adaptive EDCF: Enhanced Service Differentiation for IEEE 802.11 Wireless Ad Hoc Networks". IEEE WCNC'03 (Wireless Communications and Networking Conference), New Orleans, Louisiana, USA, March 16-20, 2003 [10] Mohammad Malli, Qiang Ni, Thierry Turletti, and Chadi Barakat, "Adaptive Fair Channel Allocation for QoS Enhancement in IEEE 802.11 Wireless LANs". IEEE ICC 2004 (International Conference on Communications), Paris, France, June 2004-9 -