active applications. In this paper we propose an admission control algorithm in ... waste AC2 and AC3 flows are allowed to utilize the available bandwidth if ...
An Admission Control Algorithm for QoS Provisioning in IEEE 802.11e EDCA Alessandro Andreadis, Giuliano Benelli, Riccardo Zambon Department of Information Engineering University of Siena Siena, Italy {andreadis,benelli,zambon}@unisi.it
Abstract—The IEEE 802.11e standard introduces QoS features for delivering multimedia applications in a WLAN environment. However, under heavy traffic loads, network saturation is easily achieved and consequently QoS performance are inevitably degraded. In this paper an admission control algorithm for resource sharing in a wireless LAN environment is proposed, with the aim at enhancing QoS support for real-time applications (e.g., VoIP and video) that are particularly sensitive to delay and jitter. The algorithm avoids saturation and protects the admitted traffics from being degraded, through a continuous monitoring of the available resources. A further optimization of the algorithm is finally implemented through the adoption of dynamic bursty transmissions times (i.e., TXOP). Performance evaluation is carried out through computer simulations, showing the benefits gained by the 802.11e EDCA scheme in terms of goodput and delay. Keywords—802.11e, admission control, QoS, EDCA, TXOP
I.
INTRODUCTION
The wide adoption of high-speed wireless connectivity has permitted the delivery of heterogeneous multimedia services almost everywhere, freeing the users from the physical constraints that are typical of wired networks. While traditional best effort applications do not pretend specific performance except reliability, multimedia applications impose more stringent requirements such as guaranteed bandwidth and bounded delays and jitter. The IEEE 802.11 wireless LAN (WLAN) standard will probably play an important role in this framework, because it is easy to install and it provides a flexible and cheap solution. However, it is well-known that this standard alone is not able to provide such Quality of Service (QoS) features; consequently, the IEEE 802.11e task group has introduced new QoS mechanisms for supporting a more efficient delivery of multimedia traffics over WLANs [1]. In particular, an enhanced MAC scheme named Enhanced Distributed Channel Access (EDCA) envisages the assignment of differentiated channel access probabilities to frames contending for channel resources. In EDCA, frame prioritization is realised through the adoption of Access Categories (AC) which are implemented in different queues at the QoS enabled stations (QSTA), being them an Access Point (QAP) or a wireless station (QSTA). ACbased prioritization is realized through independent back-off entities, by assigning to each AC different contention
978-1-4244-1653-0/08/$25.00 ©2008 IEEE
parameters (i.e., AIFSD, CWmin, CWmax, TXOPlimit) regulating channel access and frame transmission timings [1]. Even if the adoption of the new EDCA scheme provides great benefits in terms of QoS performance, when the network is heavily loaded by many traffic sources an admission control mechanism is needed in order to preserve QoS performance of active applications. In this paper we propose an admission control algorithm in support of QoS of heterogeneous multimedia traffics, which avoids network saturation and consequently guarantees good performance in terms of throughput and delay inside a wireless LAN. The admission control algorithm is based on a resource sharing scheme and on a continuous monitoring of the available resources. Their allocation is then optimized through the adoption of a TXOPlimit parameter dynamically assigned at the Access Point (AP). With this optimization, the proposed algorithm allows to overcome the fairness problems between uplink and downlink channel access times [2], with significant benefits for the global performance. Simulation results will show the substantial enhancements gained by real-time traffics (UDP video and VoIP) and how TCP traffics are not significantly unfavoured [3]. The paper is organised as follows: section II provides a description of the proposed admission control algorithm; section III describes the simulation environment, the traffics involved and the simulation results; finally, we come to the concluding remarks in section IV. II.
THE ADMISSION CONTROL ALGORITHM
The adoption of an admission control algorithm is fundamental for QoS provisioning in WLAN environments. Recent works [4][5][6] have shown that the IEEE 802.11e mechanisms for QoS support are not sufficient to guarantee QoS performance of multimedia traffics when a WLAN is overloaded. Beside AC prioritization of EDCA, a mechanism that is capable of preventing bandwidth saturation is necessary for the allocation of available resources to new traffics, without degrading the current flows. The algorithm here proposed is based on a continuous monitoring of the available channel resources in a Basic Service Set (BSS), at the purpose to avoid new traffic sources to load the cell excessively with consequent performance
ISWPC 2008
degradation of high priority flows. Hence, we propose the following resource sharing scheme for allocating the total bandwidth among the four ACs envisaged in EDCA: •
(a) 20% of bandwidth reserved to VoIP (AC0);
•
(b) 40% of bandwidth reserved to video (AC1);
•
(c) 30% of bandwidth shared between AC0 and AC1;
•
(d) 10% of bandwidth exclusively reserved to contentions between web and ftp traffics belonging to AC2 and AC3 respectively.
The above constraints impose the following logic to be always true:
⎛ Thp ref Thpeq = ⎜ ⎜ PkSz ref ⎝
⎞ ⎛ PkSz ref ⎟ × PkSz avg × 1 + log 2 ⎜ ⎟ ⎜ PkSz avg ⎠ ⎝
(1)
(2)
where Thpref is the BSS throughput (around 6Mb/s) obtained with a reference packet size PkSzref of 1024 byte. If we denote as Pktflow the number of packets sent by a traffic flow during the last observing period and as PkSzflow,, the traffic packet size for that flow, we can derive PkSzavg as follows:
∑ (Pkt × PkSz ) ∑ Pkt flow
(U AC 0 ≤ 0.5)and (U AC1 ≤ 0.7)and (U AC 0 + U AC1 ≤ 0.9)
⎞ ⎟ ⎟ ⎠
PkSzavg =
flow
flows
(3)
flow
flows
where UAC0 and UAC1 represent utilization percentages of the channel by AC0 and AC1 traffics respectively during an observing period (the observing period can be set equal to the beacon time). These sharing percentages are a good choice for a heterogeneous traffic scenario, but they can be optimized through adaptive or predictive algorithms depending on the type of traffic scenario we want to implement. It is to be noted that UDP flows (e.g., VoIP and video) behave more aggressively with respect to TCP ones (e.g., ftp and web) and also that they are usually assigned a higher EDCA priority; consequently, when there is plenty of available resources in cases a), b) and c), in order to avoid bandwidth waste AC2 and AC3 flows are allowed to utilize the available bandwidth if needed. Moreover, even if the maximum throughput of a BSS under good channel conditions can be around 6 Mb/s with big packet sizes, the WLAN global throughput is substantially reduced when the mean size of packets traversing the network diminishes, as outlined in [7]. According to this behavior, we have carried out several tests for monitoring the general trend of the 802.11b BSS total throughput versus the mean packet size, under good channel conditions (i.e., at the maximum PHY bit rate of 11 Mb/s). We have then interpolated the results and empirically obtained a formula that provides a good description on how the mean packet size affects the BSS global throughput, under the assumption that only UDP flows (i.e., voice and video) are present.
Equation (3) takes into account all flows that were active during the last observing period and also the new flow that is waiting for admission; in such a way (3) estimates PkSzavg that would result if the new flow were activated. The information on Pktflow and PkSzavg utilized by the algorithm is recovered by the TSPEC element included in the ADDTS Request frame [1]. The utilization percentages UAC0 and UAC1 can now be calculated as the ratio between the total throughput of the corresponding AC and the equivalent throughput Thpeq . Equation (2) is a good approximation in case of predominance of UDP flows, but it would be no more accurate as the number of TCP packets in the BSS becomes significantly high with respect to UDP ones. However the approximation error does not affect the performance of the proposed algorithm, due to a phenomenon of auto-regulation of AC2 and AC3 categories (i.e., web and ftp), mainly relying on TCP protocol; in fact, in case of channel saturation, the presence of UDP traffics (with higher AC priority) causes an increase of RTT and triggers TCP congestion mechanisms, thus throttling the active TCP sources (with lower AC priority).
The corresponding interpolation diagram is shown in Fig.1, where the BSS throughput values, measured at different packet sizes (i.e., 32, 64, 128, 256, 512 and 1024 bytes), are interpolated with the continuous curve representing a good empirical model for the throughput (maximum approximation error is 5%). If we define the equivalent throughput Thpeq as the BSS maximum throughput achievable with a certain mean packet size PkSzavg during an observing interval, the interpolated formula is described by the following:
Figure 1. BSS throughput vs packet size estimation.
Each time a new AC0 or AC1 flow, characterized by a certain packet size PkSzflow and target throughput Pktflow,
requires to be activated, our algorithm performs the pseudocode described in Fig.2. In this way, a source with modest throughput demands, but generating small packets, might be rejected instead of being erroneously accepted. In fact, even if its AC could be kept under its bandwidth percentage limit, such a source would contribute to reduce the average BSS packet size and the global throughput; this would be equivalent to a bandwidth reduction for all remaining flows, which could now exceed the imposed sharing limitations and provoke network saturation with QoS degradation. Figure 3. Simulation Scenario
The wireless stations generate two types of heterogeneous traffics, as seen in Table I:
- read PkSzflow - read Pktflow - estimate new PkSzavg from (3) - estimate new Thpeq from (2) - calculate UAC0 and UAC1 including the new flow - If (UAC0 > 0.5) Then reject new flow End - Else - If (UAC1 > 0.7) Then reject new flow End - Else - If [(UAC0 + UAC1)> 0.9] Then reject new flow End - Else admit new flow
TABLE I.
Figure 2. Pseudo-code of the admissin control algorithm.
In order to further enhance the algorithm and to overcome the unfairness between uplink and downlink allocation, the QAP is assigned a fine-tuned TXOPlimit value [8][9] higher than the other QSTAs, according to the following equation (valid for both AC0 and AC1):
TXOPlimit AP (i ) = log 2 N (i ) × TXOPlimitQSTA
(4)
where N(i) is the number of active stations in the BSS during the i-th beacon interval, TXOPlimitAP and TXOPlimitQSTA are the TXOPlimit values assigned at the AP and QSTA respectively. Equation (4) allows to obtain benefits in terms of uplink/downlink fairness, without increasing excessively bursty transmissions that could lead to starvation of other ACs and to the violation of the channel allocation percentages established with the admission control mechanism. III.
SIMULATIONS AND RESULTS
The simulation scenario (Fig.3) implemented in ns2 [11][12] envisages an infrastructured WLAN, with an AP compliant to the IEEE 802.11e standard, connected to a fixed server through an Ethernet switch, and eight wireless stations located in proximity of the AP (i.e., maximum 802.11b PHY bit rate). Wireless parameters follow the specification of Orinoco 802.11b 11Mbps PC card in closed environments [10].
TRAFFIC FLOWS INVOLVED IN SIMULATIONS
QSTA
Flow
AC
Bitrate (Kb/s)
Pkt size (bytes)
0 to 3
VoIP (uplink)
0
32
80
0 to 3
VoIP (downlink)
0
32
80
0 to 3
Video (downlink)
1
700
1464
0 to 3
Web (downlink)
2
---
210
4 to 7
VoIP (downlink)
0
64
160
4 to 7
VoIP (uplink)
0
64
160
4 to 7
Video (downlink)
1
300
512
4 to 7
FTP (downlink)
3
---
1500
The simulation scenario envisages the constant presence of some flows and the ingress and egress of other traffics (at each 10s interval, see Table II), in order to reach network saturation.
TABLE II.
TRAFFIC FLOWS TIMING
Flows vs time
0 to 10
10 to 20
20 to 30
30 to 40
40 to 50
50 to 60
60 to 70
70 to 80
80 to 90
90 to 100
Voip 32
6
6
8
8
8
8
8
8
4
4
Voip 64
6
6
6
8
8
8
8
8
4
4
Video 700
2
2
2
2
3
3
2
1
1
2
Video 300
1
2
2
2
2
3
3
2
2
2
Web
4
4
4
4
4
4
4
4
4
4
FTP
4
4
4
4
4
4
4
4
4
4
We carried out three types of simulations sets. The first one (named “std”) adopts the standard EDCA values of 802.11e, with no admission control. Fig.4 reports the results in terms of cumulative goodput per each AC (VoIP, video, web, ftp). We note that when the BSS is saturated,
EDCA prioritization is not sufficient for QoS support, because UDP traffics experience heavy losses and TCP flows suffer a heavy throughput decrease. The second simulation set (named “ac”, Fig.5) adopts the proposed admission control algorithm, which rejects a 700Kb/s video (at time 40s) and another at 300Kb/s (at time 50s), because they would lead to saturation, and hence the other flows are protected from performance degradation. Apparently Fig. 4 and Fig. 5 show a similar behavior, but in the first simulation more video sources are active, increasing the network load of about 1 Mb/s at time 50s; hence at this point of Fig. 4, we would expect a corresponding increase in the cumulative video throughput. On the contrary there is a decrease between 50 and 70s, which means that the saturation condition is causing heavy performance degradation (even due to more collisions).
Admission control provides significant enhancements, by protecting the active flows from being degraded by the activation of demanding traffics; these QoS enhancements are even more evident with a better-balanced TXOP, as it is clear from the stability and regularity of UDP throughputs (Fig.6). Finally, Fig. 7 reports the average delay calculated on VoIP and video traffics, in ac (admission control) and act (admission control with TXOPlimit) cases. Here, with admission control and dynamic TXOPlimit adjustment, all traffic requirements are fully respected, even if video in case 2 (ac) experiences a higher delay, with a peak of about 350 ms, that is still acceptable. It is to be noted that the delays in case 1 (std), not reported here, in case of saturation can reach much higher values that are far beyond the acceptable threshold for real-time traffics.
Figure 6. Goodput with Admission Control and dynamic TXOP (act) Figure 4. Goodput without Admission Control (std)
Figure 7. Average delay for VoIP and video in ac and act simulations Figure 5. Goodput with Admission Control (ac)
Fig. 6 reports the goodput results of a third experiment (named “act”), where a different TXOPlimit, as specified in (4), is applied to the AP with the aim to obtain new benefits in terms of uplink/downlink channel allocation balancing.
Table III reports, in the second column, the average TCP (ftp and web) throughput values, computed over the total simulation time. The third column is restricted to time interval 0-70s, corresponding to the period of maximum traffic activity;
bandwidth utilization percentages are also shown between brackets. It is evident that the adoption of admission control provides a substantial enhancement to the global throughput of ACs generally unfavoured (i.e., TCP traffics), without violating the maximum resource allocation percentages.
TABLE III.
AVERAGE TCP THROUGHPUT
Case
Throughput (Kb/s) 0-100s
Throughput (Kb/s) 0 –70s
Sim 1 (std)
639,62
227,94 (7,60%)
Sim 2 (ac)
880,55
228,06 (7,91%)
Sim 3 (act)
966,36
289,19 (9,48%)
CONCLUSIONS In this article we have proposed an admission control mechanism for QoS provisioning in IEEE 802.11e WLANs. The algorithm defines a resource-sharing scheme based on EDCA traffic prioritization, thus preventing network saturation and avoiding QoS performance degradation of sensitive flows. Simulation results carried out under simplified assumptions (i.e., good signal quality and link rate) have shown how the algorithm can significantly enhance QoS performance of multimedia flows, by dynamically adapting resource allocation according to the different traffic categories. REFERENCES [1]
IEEE Std. 802.11e-2005, Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) specifications. Amendment 8:
Medium Access Control (MAC) Quality of Service Enhancements, IEEE Std. 802.11e, 2005. [2] A. Grilo and M. Nunes, “Performance Evaluation of IEEE 802.11e”, in Proceedings of the 13th IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC), 2002, vol. 1, pp. 511-517. [3] M. Thottan and M.C. Weigle, “Impact of 802.11e EDCA on Mixed TCP-based Applications”, in Proceedings of the International Wireless Internet Conference (WICON), 2006. [4] X. Chen, H. Zhai, X. Tian and Y. Fang, “Supporting QoS in IEEE 802.11e Wireless LANs”, in IEEE Transactions on Wireless Communications, vol. 5, no 8, pp. 2217-2227, August 2006. [5] H. Zhai, J. Wang and Y. Fang, “Providing Statistical QoS Guarantee for Voice over IP in the IEEE 802.11 Wireless LANs”, IEEE Wireless Communications, vol. 13, no. 1, pp. 36-43, February 2006. [6] Y. Xiao, “QoS Guarantee and Provisioning at the Contention-Based Wireless MAC Layer in the IEEE 802.11e Wireless LANs”, IEEE Wireless Communications, vol. 13, no. 1, pp. 14-21, February 2006. [7] S. Mangold, S. Choi, P. May, o. Klein, G. Hiertz and L. Stibor, “IEEE 802.11e Wireless LAN for Quality of Service”, in Proceedings of European Wireless, 2002. [8] J. Majkowski and F.C. Palacio, “Dynamic TXOP configuration for Qos enhancement in IEEE 802.11e wireless LAN”, in Proceedings of the International Conference on Software, Telecommunications and Computer Networks (SoftCOM), 2006. [9] A. Andreadis and R. Zambon, “QoS Enhancement With Dynamic TXOP Allocation in IEEE 802.11e”, in Proceedings of the IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC), 2007. [10] W. Xiuchao, “Simulate 802.11b channel within ns2,” http://www.comp.nus.edu.sg/~wuxiucha/research/reactive/report/80211 ChannelinNS2_new.pdf, April 2004. [11] Network Simulator ns-2. http://www.isi.edu.nsnam/ns. [12] S. Wiethölter, and C. Hoene, “An IEEE 802.11e EDCA and CFB Simulation Model for ns-2”, http://www.tkn.tuberlin.de/research/802.11e_ns2/, 2006.