QoS Support for Time-Constrained Multimedia Communications in IEEE WLANs: A Performance Evaluation

Transcription

1 QoS Support for Time-Constrained Multimedia Communications in IEEE 8. WLANs: A Performance Evaluation José Villalón Francisco Micó Pedro Cuenca Luis Orozco-Barbosa Department of Computer Engineering, Universidad de Castilla-La Mancha, Albacete, SPAIN Department of Computer Engineering, Universidad de Valencia, Valencia, SPAIN [josemvillalon,pcuenca,lorozco]@info-ab.uclm.es, Francisco.M.Mico@uv.es Abstract During the past few years, we have been witnessing the massive deployment of IEEE 8. wireless LANs. Even though the widespread deployment of such wireless LANs, many studies are still underway aiming to design and develop simple yet effective QoS mechanisms for supporting time-constrained services over such platforms. In this paper, we first overview five different QoS mechanisms recently introduced in the literature as well as the upcoming IEEE 8.e () standard. We then carry out a comparative performance evaluation of all the QoS mechanisms having been reviewed. We focus our study on the effectiveness of the various QoS mechanisms for supporting time-constrained services.. Introduction Nowadays, wireless LANs are being deployed everywhere. The deployment of such networks has been spurred by the introduction of the IEEE 8. wireless LAN standard and its subsequent amendments []. Nowadays, nominal transmission rates of up to 54 Mbps are enabling the deployment of high-rate applications, such as multimedia applications. However, due to the random nature of the access channel mechanism, the provisioning of Quality of Service (QoS) support remains an open issue. Thus, the design of QoS mechanisms has become an area of active research giving rise to several new service differentiation schemes. In [], we have carried out a preliminary comparative study of various schemes and the upcoming IEEE 8.e standards. In this paper, we go a step further by examining the effectiveness of the various schemes for supporting time constrained applications. Towards this end, we consider a multiservice scenario and fix strict deadlines for the voice and video services.. Priority-Based distributed mechanisms For the sake of clarity, we first review the basic principles of operation of the IEEE 8. medium access mechanism. For further details, the reader may refer to []. The basic access function in IEEE 8. is the Distributed Coordination Function (). Under this scheme, before transmitting, a station must first sense the state of the channel. If during an interval of time, called Distributed InterFrame Space (DIFS), the channel is sensed free, the station can initiate its transmission. Otherwise, if the channel is sensed busy, once the transmission in progress finishes and in order to avoid a transmission collision, a backoff algorithm is initiated. This algorithm consists in choosing an interval of time (the backoff time) at random during which the station delays the transmission of its frame. Once having transmitted the station, it will wait to get back a reply from the destination station. If after waiting for a time interval, denominated Short InterFrame Space (SIFS<DIFS), the station does not get an Immediate Positive Acknowledgement (ACK) from the destination station, it simply assumes that there has been a collision. The station can then attempt to retransmit a finite number of times using a longer backoff time after each attempt. This access function is easy to implement and suitable for most applications. However, it does not provide Quality of Service (QoS) support. This work was supported by the Ministry of Science and Technology of Spain under CICYT project TIC3-854-C6-, the Council of Science and Technology of Castilla-La Mancha under project PBC-3- and FEDER.

2 .. et al. mechanism The proposed mechanism [3] uses two properties of the IEEE 8. standard for providing various priority services: the Interframe Space (IFS) used between data frames and the backoff mechanism. By making use of two different time length for IFS, the stations implementing the shortest IFS will be assigned a higher priority than a station with a longer IFS. By using these two different interframe spaces (PIFS and DIFS), traffic can be differentiated and classified into two classes. To further extend the number of available classes, the backoff mechanism could be used to differentiate between stations. This can be simply achieved by defining several time lengths for the backoff interval used in the backoff algorithm depending on the priority level assigned to the station. Table. priority classes Priority IFS Algorithm 3 PIFS Rand() x +i / PIFS +i / + Rand() x +i / DIFS Rand() x +i / DIFS +i / + Rand() x +i / Table I shows four different priority classes (-3) defined using two different interframe spaces, and two different backoff algorithm [3]. The backoff algorithm guarantees that the low priority stations always generate longer backoff intervals than the stations with higher priority... Aad et al. mechanism. Aad et al. present in [4] three differentiation mechanisms... Different backoff increase function (referred from now on as Add- scheme). Each priority level has a different backoff increment function. A shorter contention window is assigned to the higher priority stations ensuring that in most (although not all) cases, high-priority stations are more likely to access the channel than the low-priority ones. This method modifies the contention window of the priority level j after i transmission attempts as follows: CW = P j x CW, where P j is a factor used to achieve service differentiation with the highest value for lowest priority stations. In each retry, a higher increment factor implies a higher waiting time and therefore a smaller throughput.... Different backoff increase function (referred from now on as Aad-CWmin scheme). The main motivation is that, with a small number of stations contending for the access to the channel, most of the time CW remains at its minimum value (CWmin). Therefore, an Add-backoff scheme will not help to promptly resolve access conflicts as the CWs will rarely increase. This led to the authors to use different values for CWmin. This scheme consists on defining a different CWmin j value per priority level. Assigning the shortest CWmin to the highest priority level ensures that in most (although not all) cases, the highestpriority are more likely to first gain access to the channel than the lower-priority stations...3. Different DIFS (referred from now on as Add-DIFS scheme). Each station uses a different DIFS corresponding to its priority level. Under this approach, each priority level uses a different DIFS, for example, DIFS j+ < DIFS j. Before transmitting a packet, the stations having priority j + will wait for an idle period of length DIFS j+ slot times. To avoid collision between frames within a same priority, the backoff mechanism is maintained in a way that the maximum contention window size added to DIFS j is DIFS j- -DIFS j. This ensures that no station of priority j has queued frames before station of priority j- starts transmitting. This may lead to starvation if a station with a high priority always has frames to transmit, since none of the stations with lower priority will ever be able to access the channel..3. (Tiered Contention Multiple Access) mechanism This mechanism by Benveniste [5] defines the Arbitration-time InterFrame Space (AIFS) as the mandatory waiting time before a transmission can be initiated. It also suggests the use in the backoff algorithm of the -Counter Update Time (BCUT) regulating the decrement rate of the backoff counter (aslottime in ). The priority classes are set by making use of several parameters. First, different AIFS and BCUT values are assigned to each class (a smaller AIFS allows the station to promptly access the channel, while a smaller BCUT enables a faster decrement of the backoff counter). Second, as in the Aad- mechanism, different Persistence Factors (PF) are assigned. In the case of stations with delay sensitive traffic, the backoff time diminishes instead of increasing exponentially like in. Finally, a deadline is set for the time that a frame stays in the MAC layer waiting to be transmitted, and therefore a frame surpassing this time limit is

3 discarded. Since PCF accesses the channel after an idle period of length PIFS, assigning the top priority class an IFS=PIFS and a minimum backoff delay is equal to will avoid conflict with the centralized protocol. 3. The upcoming IEEE 8.e standard The IEEE 8.e draft [6] is a proposal defining the QoS mechanisms for wireless LANs for to supporting time-sensitive applications such as voice and video communications. IEEE 8.e incorporates two new access mechanisms: the contention-based Enhanced Distributed Channel Access (), known in the previous drafts as the Enhanced (E) and the HCF Controlled Channel Access (HCCA). One main feature of IEEE 8.e is the definition of four Access Categories (AC) queues and eight User Priorities (UP) queues at MAC layer. Another main feature is the concept of Transmission Opportunity (TXOP), which defines the transmission holding time for each station. (AIFSN) (CWmin) (CWmax) 8.e: up to 8 User Priorities (UPs) per QSTA 8 UPs mapping to 4 Access Categories (ACs) AC AC AC AC3 (AIFSN) (CWmin) (CWmax) (AIFSN) (CWmin) (CWmax) (AIFSN3) (CWmin3) (CWmax3) scheduler (resolves virtual collisions by granting TXOP to highest priority Tranmission attempt Figure. Access has been designed to be used with the contention-based prioritized QoS support mechanisms. In, two main methods are introduced to support service differentiation. The first one is to use different IFS values for different ACs. The second method consists in allocating different CW sizes to the different ACs. Each AC forms an independent entity with its own queue and its own access mechanism based on with its own Arbitration Inter-Frame Space ( AIFS [ AC] = SIFS + AIFSN[ AC] SlotTime ) and its own CW[AC] (CWmin[AC] CW[AC] CWmax[AC]) (see figure ). If an internal collision arises among the queues within the same QSTA, the one having higher priority obtains the right to transmit. It is said that the queue that is able to gain access to the channel obtains a transmission opportunity. Each TXOP has a limited duration (TXOPLimit) during which an AC can send all the frames it wants. 4. Performance evaluation In this section, we carry out a comparative study of five different Priority-Based Distributed Mechanisms (namely,, Aad-, Aad-CWmin, Aad-DIFS, and mechanisms) and the IEEE 8.e () upcoming standard. Throughout our studies, we have considered the use of a WLAN supporting four different types of services: voice, video, besteffort and background traffic. The QoS provided by all these schemes have been evaluated in terms of five different metrics: throughput, collision rate, access delay, delay distribution and packet loss rate. In this section, we study the effect on the performance of the system when delay bounds of voice and video applications are applied. For all our studies, we have made use of the OPNET Modeler tool. [7]. In our simulations, we model an Mbit/s IEEE 8.b wireless LAN supporting four types of services: Voice(Vo), Video(Vi), Best-effort(BE) and Background(BK). We assume the use of a wireless LAN consisting of several wireless stations and an access point connected to a wired node that serves as sink for the flows from the wireless domain. Each wireless station transmits a single traffic type (Vo, Vi, BE or BK) to the access point. We assume the use of constant bit-rate voice sources encoded at a rate of 6 kbits/s according to the G.78 standard. For the video applications, we have made use of the traces generated from a variable bit-rate H.64 video encoder[8]. The average video transmission rate is around 48 kbits/s. The best-effort and background traffics have been created using a Pareto distribution traffic model with average sending rate of 8 kbit/s and 56 kbit/s, respectively. For all the scenarios, we have assumed that one fourth of the stations support one of the four kinds of services: voice, video, BE and BK applications. We start by simulating a WLAN consisting of four wireless stations (each one supporting a different type of traffic). We then gradually increase the Total Offered Load of the wireless LAN by increasing the number of stations by four. We increase the number of stations 4 by 4 starting from 4 and up to 36. In this way, the normalized offered load is increased from. up to.. We have preferred to evaluate a normalized offered load, rather than the absolute value. The normalized offered load is determined with respect to the theoretical maximum capacity of the Mbit/s IEEE 8.b mode, i.e. 7. Mbit/s. In order to be able to make a fair comparison, we have used the parameter setting of the priority-based distributed mechanisms (table ) as close as possible to the settings used in [9]. In order to limit the delay experienced by the video and voice applications, the maximum time that video

4 packet and voice packet may remain in the transmission buffer has been set to ms and ms, respectively. These time limits are on-line with the values specified by the standards and in the literature. Whenever a video or voice packet exceeds these upper bounds, it is dropped. The loss rate due to this mechanism is given by the packet loss rate due to deadline. Aad DIFS Aad Aad CWmin TCM A Table. Parameter settings AC IFS CWmin CWmax PF Vo x Slot_time + SIFS 3 3 Vi x Slot_time + SIFS 3 3 Be x Slot_time + SIFS 3 3 Bk x Slot_time + SIFS 3 3 Vo x Slot_time + SIFS 3 3 Vi x Slot_time + SIFS 3 3 Be 4 x Slot_time + SIFS 3 3 Bk 7 x Slot_time + SIFS 3 3 Vo x Slot_time + SIFS 3 3 Vi x Slot_time + SIFS 3 3 Be x Slot_time + SIFS Bk x Slot_time + SIFS Vo x Slot_time + SIFS 7 3 Vi x Slot_time + SIFS 5 3 Be x Slot_time + SIFS 3 3 Bk x Slot_time + SIFS 3 3 Vo Slot_time + SIFS 3 3 Vi x Slot_time + SIFS 3 3 Be 4 x Slot_time + SIFS Bk 8 x Slot_time + SIFS Vo x Slot_time + SIFS 7 5 Vi x Slot_time + SIFS 5 3 Be 4 x Slot_time + SIFS 3 3 Bk 7 x Slot_time + SIFS 3 3 AC IFS Algorithm Vo Slot_time + SIFS Rand() x +i / Vi Slot_time + SIFS +i / + Rand() x +i / Be x Slot_time + SIFS Rand() x +i / Bk x Slot_time + SIFS +i / + Rand() x +i / Figure shows that for mild loads as low as.5, all the mechanisms, except and, are unable to avoid the loss of voice packets resulting into the degradation of the normalized throughput. When the load reaches.7, the performance of also degrades while is able to deliver the entire submitted load. This is due to the higher collision rate of the former. It is also important to mention that the deadline imposed by the voice application is shorter than the mean access delay in the presence of retransmission attempts due to collisions. The results obtained by the and schemes are even worse mainly due to the even higher collision rate exhibited by these schemes than for any other scheme. Figure shows similar results for the video traffic. All the mechanisms, except and, start to lose video packets for loads as low as.5. exhibits better performance than for loads higher than.85. From all the other schemes, Aad- and Aad-DIFS show the worst results. The former shows the effectiveness of incrementing the window size at different rates for the various traffic types. The latter shows that it is also possible to differentiate the traffic by making use of different values of IFS. In the case of the BE and Bk, these are severely affected as the network load is increased. This is in line with the specifications of the various QoS-aware mechanisms by guaranteeing the QoS to the voice and video services. As expected, shows the best results for this type of traffic while the other schemes limit the resources available to these flows in their attempt to offer QoS guarantees to the voice and video traffic. Once again, the scheme proposed by offers the worst results despite being unable to offer QoS guarantees to the time-constrained traffic (voice and video) Voice traffic Best Effort traffic Background traffic.4. Figure. Average Throughput Total Throughput Total Figure 3. Total Throughput and retransmissions Figure 3 shows the overall throughput and packet retransmissions per packet for all the services under study. Figure 3 clearly shows that the lowest number of retransmissions is obtained when making use of the scheme. This in turn translates into a higher throughput. This shows the effectiveness of incrementing the IFS and backoff times at a different rate for the various types of traffic. On the contrary, the scheme proposed by exhibits a higher number of collisions due to the short window size employed by this scheme. Furthermore, this scheme uses the same

5 backoff mechanism for all service classes. Regarding the scheme, this one exhibits a good performance for loads of up to.75. However, beyond this point, the number of collisions increases rapidly mainly due to the CW max used for the voice and video services, 5 and 3, respectively Voice traffic Best Effort traffic Background traffic Figure 4. Average number of attempts per packet Figure 4 shows the mean number of packet retransmissions for each type of service. All of them show similar behavior to the global number of retransmissions. In the case of the mechanism, it is clear that the decrease on the voice normalized throughput as the offered load reaches.75 is due to the repeated number of packet retransmissions. The same can be said for the video traffic. The scheme exhibits much better performance since the retransmission rate is two to three times lower than the one for the scheme. This also has a direct impact on the mean access delay. Figure 5 shows the mean access delay per service class. For the voice traffic, the mechanism offers the lowest delay independently on the number of workstations due to shorter waiting times (IFS, CW). Similar results are obtained for the video traffic. Up to loads of.9, offers the lowest delay. For higher loads, due to the increasing number of collisions, the performance of degrades rapidly. In turn the scheme offers better results for loads higher than.9. Once again, the algorithm proposed by exhibits the worst performance due to the high collision rate. From Figure 5, it is clear that the schemes offering the best results for the voice and video traffic heavily penalize the best-effort traffic. 4 x Voice traffic 3 3 Best Effort traffic Background traffic 4 Figure 5. Average access delay Voice: Probability [delay <= x] ( =.744) BE: Probability [delay <= x] ( =.744) Video: Probability [delay <= x] ( =.744) BK: Probability [delay <= x] ( =.744) Figure 6. Cumulative distribution () of the access delays Figure 6 shows the cumulative distribution function of the access delay for all mechanism operating at a load close to.75. This metric better shows the access delay offer by the different mechanisms. From the figures, it is clear that the scheme offers the lowest access delays for the voice and video traffic. Up to this load level,.75, the scheme is the one offering the best guarantees. Figure 7 depict the packet loss rate due to the missing of the transmission deadline for the voice and video traffic services, respectively. These results help us to better understand the degradation on the normalized voice and video throughput shown in Figure. From Figure 7, it is clear that the scheme exhibits the lowest loss rate due to collisions.

6 Packet Loss Rate Voice traffic Packet Loss Rate Figure 7. Packet loss rate due to deadline It is also worth to mention that in an IEEE 8. LAN, losses may also occur when the maximum number of allowed retransmissions attempts is reached. In our study, only.% of the BE and BK packets have experienced such kind of losses. While in the case of voice and video traffic, no losses due to this cause have been reported. 5. Conclusions In this paper, we have evaluated various QoS provisioning schemes as well as the IEEE8.e. We have considered a scenario consisting of a wireless LANs where the stations support four different types of services: voice, video, best-effort and background. Throughout an exhaustive campaign of simulations, we have evaluated the performance of the system in terms of five metrics: throughput, collision rate, access delay, delay distribution and packet loss rate. We have further analyzed the various sources giving rise to packet losses, such as, losses due to violation of the timing requirements of real-time applications. We have studied the effect on the performance of the system when these delay bounds of voice and video applications are applied. Our results show that the network degrades badly as the number of collisions increases. The performance of the time-constrained applications, in particular voice, degrades even further due to the short deadline imposed by this type of applications. The impact over the network performance is clear on the scheme proposed by. This scheme exhibits the worst performance results. Similarly, the is unable to guarantee a good performance for loads beyond.75. In this latter scheme, the steep performance drop is mainly due to the excessive number of collisions. The collisions are in turn mainly due to the fact that the IFS parameter has been fixed to the same value for the video and voice services. Furthermore, the values used for CWmax are too short (5 and 3) contributing to a higher collision probability. Regarding the scheme, this is the one showing the best results among the schemes under consideration. This scheme makes use of different values for the IFS and PF parameters. It also uses higher values for the CWmax. From our results, we can conclude that the scheme does not perform well when exposed to heavy loads mainly due to the excessive number of collisions. The performance of can be considerable improved by properly tuning its parameters. References [] LAN MAN Standards Committee of the IEEE Computer Society, ANSI/IEEE Std 8., Part : Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications, 999 Edition. [] J. Villalón, P. Cuenca, L. Orozco-Barbosa, QoS Provisioning Mechanisms for IEEE 8. WLAN: A Performance Evaluation, th IFIP International Conference on Personal Wirless Communications, Colmar, France, August 5 (to be presented) [3] D.J. and R.S. Chang, A priority scheme for IEEE 8. access method, IEICE Trans. Commun., vol. E8-B, num., pp. 96-, January 999. [4] I. Aad and C. Castelluccia, Priorities in WLANs, Computer networks, vol. 4/4, pp , February 3. [5] M. Benveniste, "'Tiered contention multiple access' (), a QoS-based distributed MAC protocol", The 3th IEEE International Symposium on Personal, Indoor and Mobile Radio Communications, vol., pp , September. [6] IEEE 8 Committee of the IEEE Computer Society, IEEE P8.e/D. Draft Amendment to IEEE Std 8., Part : Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications: Medium Access Control (MAC) Quality of Service (QoS) Enhancements, October 4. [7] Opnet.Technologies.Inc. OPNET Modeler [8] ITU-T Recommendation H.64. Advanced Video Coding For Generic Audiovisual Services. May 3 [9] G. Chesson, W. Diepstraten, D. Kitchin, H. Teunissen and M. Wentink, Baseline D-QoS Proposal. IEEE 8. working group document 8.-/399 ().