1. 12/13/2006 Improving Quality of Service for VoIP Traffic in IEEE 802.11 Wireless Networks Sangho Shin Henning Schulzrinne

2. Introduction Increased Usage of VoIP service over wireless networks Widely deployed WLANs Shopping Malls Coffee shops Streets Parks Wi-Fi phones New service plans Sprint – cellular and Wi-Fi Limited QoS support in 802.11 WLANs IEEE 802.11e ? [skyhook] Map of APs in Manhattan

3. Outline Overview of QoS problems My earlier work for the problems Fair resource distribution b/w uplink and downlink using APC Call Admission Control with QP-CAT Conclusion Future work

4. Framework of VoIP over WLANs AP Router 160.38.x.x128.59.x.x Internet PBX Wireless client Introduction

5. QoS problems of VoIP service The network disruption during L2/L3 handoff Fast L2/L3 handoff Limited VoIP capacity Dynamic PCF (DPCF) Unbalanced uplink and downlink delay Adaptive Priority Control (APC) Significant deterioration of QoS in the case of channel congestion Call admission control using QP-CAT Introduction

6. Outline Overview of QoS problems My earlier work for the problems Fair resource distribution b/w uplink and downlink (APC) Call Admission Control with QP-CAT. Conclusion Future work Seamless layer-2 handoff Fast layer-3 handoff Dynamic PCF

7. Fast layer-2 handoff [26] (1/2) Earlier work AP Router Wireless clientAll APs New AP Authentication request Authentication response Association request Association response Probe request (broadcast) Probe responses AP L2 handoff trigger Authentication delay Probe delay Association delay 300m ~ 1s 2 ms [26] Sangho Shin, Andrea G. Forte, Anshuman Singh Rawat, and Henning Schulzrinne. Reducing MAC layer handoff latency in IEEE 802.11 wireless LANs. In ACM MobiWac '04, pages 19~26, New York, NY

8. ms Fast layer-2 handoff (2/2) Selective Scanning Scan non-overlapping channels first Channel 1, 6, and 11 in 802.11b Reduces the scanning time to 1/3 Caching Motivated by locality Store the scanned AP information in a cache Use it in the future handoffs Handoff without scanning Reduces the total handoff time to 4 ms No changes in the infrastructure  Practical solution! Earlier work Experiments in 802.11b WLANs

9. Fast layer-3 handoff [9] (1/2) Earlier work AP Router 160.38.x.x128.59.x.x L2 handoff L3 handoff Subnet change detection A few minutes New IP acquisition- DHCP 1 second DHCP Discover DHCP Offer DHCP Request DHCP ACK DAD DHCP procedure [9] Andrea Forte, Sangho Shin, and Henning Schulzrinne. Improving layer 3 handoff delay in IEEE 802.11 wireless networks. In WICON, Aug 2006. DAD: Duplicate Address Detection

10. Total handoff time Experiments in 802.11b ms Fast layer-3 handoff (2/2) Fast subnet change detection Broadcast a bogus DHCP request The DHCP server responds with DHCP NACK Check the IP address of the DHCP server Temporary IP address Scan potentially unused IP addresses in the new subnet Transmit multiple ARP packets Pick a non-responded IP address as a temporary IP address Use it until a new IP address is assigned by the DHCP server Revisits the previous subnets IP address lease not expired  Use the old IP address! Earlier work ARP DHCP Discover timeout Update sessions DHCP Offer DHCP Request Update sessions DHCP Ack DHCP Request DHCP NACK Subnet change L2handoff 180 30

11. [18] Takehiro Kawata, Sangho Shin, and Andrea G. Forte. Using dynamic PCF to improve the capacity for VoIP traffic in IEEE 802.11 networks. In WCNC, IEEE, vol 3, pages 13~17, Mar 2005. Dynamic PCF (DPCF) [18] (1/2) PCF (Point Coordination Function) Polling based media access No contention, no collision Polling overhead No data to transmit  Unnecessary polls waste bandwidth Big overhead, considering the small VoIP packet size. Earlier work Contention Free Period (CFP) Contention Period (CP) Contention Free Repetition Interval DCF poll Poll+datapoll BeaconCF-End data Null Polling overhead

12. Dynamic Polling List Store only “active” nodes Higher priority for VoIP Simulation results in 802.11b 0123 Number of Data Sessions Delay (90th%tile) and throughput of 28 VoIP sources and data traffic 0 500 1000 1500 2000 2500 3000 Throughput (kb/s) 0 100 200 300 400 500 600 End-to-End Delay (ms) Dynamic PCF (DPCF) (2/2) CFPCP Polling List 12345678 1 38 65 Null ACK 1 234 78 7 5 6 7 Queue CFPCP Dynamic Polling List 138 1 3 8 123 65 ACK 7 Earlier work DCF 25 400 487 545 DPCF 11 22 26 29 PCF 18 67 183 312 Data throughput VoIP throughput

13. Outline Overview of QoS problems My earlier work for the problems Fair resource distribution b/w uplink and downlink using APC in DCF Call Admission Control with QP-CAT Conclusion Future work Motivation Adaptive Priority Control Simulation results Related work Conclusion

14. Motivation Big gap between uplink and downlink delay Why? - DCF The same chance to transmit frames among nodes In VoIP traffic, the AP has more packets to transmit than each client APC Solution ? Give a higher priority to the AP DCF = Distributed Coordination Function 64kb/s VBR VoIP traffic 802.11b 11 Mb/s via simulations

15. Control Contention Window (CW) Shorter CW  smaller backoff time  higher priority Hard to control Backoff = random (0, CW) Higher collision rate Control Inter-Frame Spacing (IFS) Increase IFS  lower priority Not accurate because of BO Increase the delay due to the increased IFS How to control the priority? (1/2) APC Node A Node B backoff Node B Node A Tx in DCF BOframe DIFS Channel DIFS = DCF Inter-Frame Spacing

16. Priority of node A is 3 times of node B How to control the priority? (2/2) Contention Free Transmission (CFT) Transmit frames w/o backoff Control the number of frames to be sent contention free Accurate priority control No overhead Reduce the overall transmission time  Improve the capacity APC Node A Node B

17. Optimal priority of the AP (P) Intuitive method P  the number of wireless clients (N) Adapts to change in the number of VoIP sources Not adapts to change in the traffic volume APC P  Q AP /Q Nodes Q AP is the number of packets in the queue of the AP Q Nodes is the average number of packets in the queue all nodes Adapts to instant change of uplink and downlink traffic load APC

18. APC – example 1 AP DS queue APC Q AP = 4 Q Node = 1 P = 4/1 = 4

19. APC – example 2 AP DS queue APC Q AP = 4 Q Node = 2 P = 4/2 = 2

20. APC – example 2 AP DS queue APC Q AP = 2 Q Node = 1 P = 2/1 = 2

21. Simulation results of APC (1/3) APC Threshold Capacity DCF Threshold Capacity APC 28 Calls  35 calls (25%) 802.11b 11Mb/s 64kb/s VBR traffic 20ms pkt intvl 0.39 activity ratio

22. Simulation results of APC (2/3) APC 90th%tile delay of VoIP traffic (35 calls)

23. Simulation results of APC (3/3) 10ms + 20ms Packetization Interval20ms + 40ms Packetization Interval APC

24. Related work Many papers about fairness in throughput ([Deng et. al. IEICE Trans. Comm.], [Barry et. al. Infocom ’01], [Aad Infocom ’01], [Tickoo et. al. Infocom ’04]) In VoIP traffic, Jain’s Fairness Index is almost 1 regardless of the big gap Wang et al. [AINA ’04] - New Fair MAC Fairness between clients Clients transmit all packets consecutively at most for Maximum Transmission Time (MTT) Improved only a few ms  low uplink delay Casetti et al. [PIMRC ’04] Based on EDCA Found a fixed optimal CW by simulations Improved the capacity by 15% Not a global solution for all traffic types APC

25. Conclusion Uplink and downlink delay of VoIP traffic in DCF are significantly unbalanced  Unfair resource distribution b/w uplink and downlink APC balances the uplink and downlink delay very well by allowing the AP to transmit Q AP /Q Nodes packets using Contention Free Transmission (CFT) APC improves the capacity for VoIP traffic from 28 calls to 35 calls, by 25% APC

26. Future work Implement APC in 802.11e and evaluate it with various background traffic Implement APC using the Mad-Wifi driver and evaluate the performance in the OBRIT test-bed Improve APC so that it does not require client information APC

27. Outline Overview of QoS problems My earlier work for the problems Fair resource distribution b/w uplink and downlink using APC Call Admission Control with QP-CAT Conclusion Future work Motivation and requirements Metric for admission control Queue size Prediction using CAT Simulation results Related work Conclusion

28. Motivation When channel is congested, delay of all VoIP flows significantly increases Experimental results in the ORBIT test-bed 64kb/s CBR VoIP traffic 802.11b 11Mb/s 20ms packet interval

29. Requirements for CAC Accurate (Guarantee QoS of existing flows) Need a good metric for QoS Fast Users should not wait for a long time to call Efficient Minimum waste of bandwidth Flexible (extensible) Many types of VoIP traffic need to be supported Admission Control using QP-CAT

30. Metric (1/3) Queue size of the AP (The number of packets in the queue of the AP) Correlation b/w the queue size and downlink Admission Control using QP-CAT Experimental results in the ORBIT test-bed 802.11b 11 Mb/s 64 kb/s CBR 20 ms Pkt Intvl

31. Metric (2/3) Theoretical model Downlink delay Queuing delay Transmission delay Processing time of the AP Queue size of the AP Admission Control using QP-CAT According to computation using 802.11b parameters

32. Errors according to the delay Cumulative Distribution Function of errors Metric (3/3) Errors of the model Admission Control using QP-CAT

33. Queue size Prediction (QP) How to use the metric ? When the queue size goes beyond a threshold, then reject the future calls ? Too late  QoS already deteriorated Cannot disconnect already admitted flows See the future! Need to predict the queue size in advance Admission Control using QP-CAT

34. Computation of Additional Transmission (CAT) (1/5) Basic concepts Emulate the additional VoIP flows  Predict the queue size Emulate VoIP traffic Packets from a new flow Compute Additional Transmission channel Actual packets Additional transmission Decrease the queue size Predict the future queue size + current packets additional packets

35. CAT (2/5) Emulation of VoIP flows Two counters: DnCounter, UpCounter Follow the same behavior of VoIP flows Increase the counters every packetization interval of the flows Decrement the counters alternatively 20ms time DnCounter++ UpCounter++ DnCounter++ UpCounter++ DnCounter++ UpCounter++ 20ms Example : 20ms packetization interval Admission Control using QP-CAT

36. CAT (3/5) Computation of transmission time of a VoIP frame (T t ) DIFS backoff TvTv SIFS ACK frame VoIP packet TbTb T ACK TtTt PLCPMACIPUDPRTPVoice data T t = DIFS + T b + T v + SIFS + T ACK Admission Control using QP-CAT

37. CAT (4/5) Computing Additional Transmission (n p ) T c = t 2 - t 1 DnCounter--UpCounter-- Busy medium TtTt TtTt TrTr Admission Control using QP-CAT t1t1 t2t2

38. CAT (5/5) Additional considerations Collision Deferral Serialization of downlink packets

39. 16 calls 17 calls + 1 call 16 calls + 1 call 17 calls 17 Calls18 Calls Simulation results (1/3) Another call is added 32 kb/s VoIP traffic with 20ms packetization interval

40. 14 calls + 1 call15 calls + 1 call 14 calls15 calls 16 calls Simulation results (2/3) 64 kb/s VoIP traffic with 20ms packetization interval

41. 29 calls + 1 call30 calls + 1 call 29 calls30 calls 31 calls Simulation results (3/3) 32 kb/s VoIP traffic with 40ms packetization interval

42. Related work (1/3) Numerical or theoretical approaches Yang Xiao et al. [Communication Magazine ‘04 ] 802.11e EDCA based access control Compute available bandwidth using TXOPs and announce it to clients Guarantee bandwidth, but not delay  Applicable Video traffic only Pong et al. [Globecom’03] Estimate available bandwidth using an analytical model Check if the requested BW available by changing CW/TXOP The assumption of the analytical model is far from real environment Kuo et al [Globecom’03] Pure analytical model based Expected bandwidth and delay are computed using an analytical model

43. Related work (2/3) CBR/CUE - Sachin et al. [Globecom’03] and Zhai et al. [QShine04 ] New metric : Channel Utilization Estimate (CUE)/Channel Business Ratio (CBR) = fraction of time per time unit needed to transmit the flow CUE per flow is computed using the average Tx rate of each flow Clients compute the CBR from their Tx rate and send it to the AP regularly. Compare the remaining CUE and the requested CUE Assume 15% of wasted bandwidth due to collision or fluctuation  0.85 max total CUE Actual probing Metric: delay and packet loss Used for wired networks Very accurate and simple Waste a certain amount of bandwidth

44. Related work (3/3) Comparison ApproachesMetricAssumptionAdapts to channel Waste of BWExtensibility Theoretical approaches CW/TXOP Computed bandwidth Saturated channel NoLowGood CUE/CBRCUE CBR Max CU (Measured in advance) No (Fixed Max CU) Middle (Reserved BW for collisions) Good Actual Probing Delay/ packet loss NoYesHigh (Probing flow) Bad QP-CATQueue size of the AP NoYesLowGood

45. Conclusion The queue size of the AP is a good indicator for QoS of all existing calls QP-CAT can predict the future queue size of the AP very well We can perform call admission control using QP-CAT accurately and efficiently

46. Future work Evaluate the QP-CAT using various types of background traffic Implement call admission control framework using QP-CAT Evaluate the efficiency of CAC with QP- CAT using actual call arrival rate and call duration

47. Contributions APC Introduced a New dynamic priority control method, CFT and showed it is better than others (CW, IFS). Verified that the optimal priority of the AP for balancing uplink and downlink delay is Q AP /Q Node, using CFT. Increased the VoIP capacity by 25% QP-CAT Verified the linear correlation between the number of packets in the queue of the AP and downlink delay of VoIP traffic via experiments Enabled the AP to predict the future downlink delay accurately using CAT Protect the QoS of the existing VoIP traffic in fluctuating channel conditions, minimizing the waste of bandwidth

48. Timetable TimelineWorkProgress ~ Dec. 2006Implementation in the QualNet simulator and evaluation of APC and QP-CAT Completed Jan. 2007 ~Apply QP-CAT to call admission control and evaluate the efficiency Mar. 2007 ~Evaluate the call admission control with QP- CAT with various types of background traffic Apr. 2007 ~Integrate APC with 802.11e using the QualNet simulator May. 2007 ~Evaluate APC with various types of background traffic Jun. 2007 ~Implement APC with the MadWifi driver and evaluate it in the ORBIT test-bed Aug. 2007 ~Introduce and implement a new APC that does not use the queue size of nodes Oct. 2007 ~Start to write my thesis Feb. 2008Defend my thesis

49. Thank you

50. References (1/3) [1] W. Arbaugh A. Mishra, M. Shin. An Empirical Analysis of the IEEE 802.11MAC Layer Handoff Process. ACM SIGCOMM Computer Communication Review, 33(2):93~102, April 2003. [2] I Aad and C Castelluccia. Differentiation mechanism for IEEE 802.11. In IEEE INFOCOM, pages 209~218, Apr 2001. [3] N. Akhtar, M. Georgiades, C. Politis, and R. Tafazolli. SIP-based end system mobility solution for all-IP infrastructures. In IST Mobile & Wireless Comm. Summit 2003, June 2003. [4] M Barry, A T Campbell, and A Veres. Distributed control algorithms for service differentiation in wireless packet networks. In IEEE INFOCOM, pages 582~590, Apr 2001. [5] G. Camarillo, W. Marshall, and J. Rosenberg. Integration of Resource Management and Session Initiation Protocol (SIP). RFC 3312, IETF, Oct 2002. [6] Casetti, C. Chiasserini, and C.F. Improving fairness and throughput for voice traffic in 802.11e EDCA. PIMRC 2004. vol 1, pages 525~530, 2004. [7] D J Deng, R S Chang, and A Veres. A priority scheme for IEEE 802.11 DCF access method. IEICE Trans. Commun., E82-B(1):96 ~ 102, Oct 1999. [8] R. E. Droms. Dynamic Host Conguration Protocol (DHCP). RFC 2131, Internet Engineering Task Force, March 1997. [9] Andrea Forte, Sangho Shin, and Henning Schulzrinne. Improving layer 3 handoff delay in IEEE 802.11 wireless networks. In WICON 2006, Aug 2006. [10] Sachin Garg and M. Kappes. Admission control for VoIP trafc in IEEE 802.11 networks. In GLOBECOM, pages 3514~3518, Dec 2003.

51. References (2/3) [11] IEEE. Wireless LAN Medium Access Control (MAC) and Physical (PHY) specifications, 1999. [12] IEEE. Wireless LAN Medium Access Control (MAC) and Physical (PHY) specifications: Higher-Speed Physical Layer Extension in the 2.4GHz Band, 1999. [13] IEEE. IEEE Draft Std. 802.11e, Medium Access Control (MAC) Enhancements for Quality of Service (QoS), D8.0 edition, Feb 2004. [14] ITU-T G.114. One-way Transmission Time, 2003. [15] ITU-T P.59. Artificial Conversational Speech, 1993. [16] R Jain, D Chiu, and W Hawe. A quantitative measure of fairness and discrimination for resource allocation in shared computer systems. Technical Report TR-301, DEC, 1984. [17] Jiwoong Jeong, Sunghyun Choi, and Chong kwon Kim. Achieving weighted fairness between uplink and downlink in IEEE 802.11 DCF-based WLANs. In Qshine, August 2005. [18] Takehiro Kawata, Sangho Shin, and Andrea G. Forte. Using dynamic PCF to improve the capacity for VoIP traffic in IEEE 802.11 networks. In WCNC, 2005 IEEE, vol 3, pages 13~17, Mar 2005. [19] Sung Won Kim, Byung-Seo Kim, and Yuguang Fang. Downlink and uplink resource allocation in IEEE 802.11 wireless LANs. IEEE Trans. on Vehicular Technology, 54(1):320~327, Jan. 2005. [20] Yu-Liang Kuo, Chi-Hung Lu, E.H.K.Wu, and Gen-Huey Chen. An admission control strategy for differentiated services in IEEE 802.11. In GLOBECOM, pages 707~712, Dec 2003.

52. References (3/3) [21] O.Tickoo and B.Sikdar. Queueing analysis and delay mitigation in IEEE 802.11 random access MAC based wireless networks. In INFOCOM, March 2004. [22] S Pilosof, R Ramjee, D Raz, Y Shavitt, and P Sinha. Understanding TCP fairness over wireless LAN. In IEEE INFOCOM, pages 863~872, Mar 2003. [23] Dennis Pong and Tim Moors. Call admission control for IEEE 802.11 contention access mechanism. In GLOBECOM, pages 174–178, Dec 2003. [24] QualNet Network Simulator. [25] Henning Schulzrinne, S. Casner, R. Frederick, and V. Jacobson. Rtp: A transport protocol for real-time applications. RFC 3550, IETF, Jul 2003. [26] Sangho Shin, Andrea G. Forte, Anshuman Singh Rawat, and Henning Schulzrinne. Reducing MAC layer handoff latency in IEEE 802.11 wireless LANs. In ACM MobiWac '04, pages 19~26, New York, NY, USA, 2004. ACM Press. [27] TIA. Voice Quality Recommendations for IP Telephony, 2003. [28] Xin Gang Wang, Geyong Min, Mellor, and J.E. Improving VoIP application's performance over WLAN using a new distributed fair MAC scheme. In Advanced Information Networking and Applications, 2004. AINA 2004. 18th International Conference, volume 1, pages 126~131, 2004. [29] Yang Xiao and Haizhon Li. Evaluation of distributed admission control for the IEEE 802.11e EDCA. Communications Magazine, IEEE, 42(9):S20~S24, Sept 2004. [30] Hongqiang Zhai, Xiang Chen, and Yuguang Fang. A call admission and rate control scheme for multimedia support over IEEE 802.11 wireless LANs. In Qshine, pages 76~83, Jan 2004.