1. A Credit-based Home Access Point (CHAP) to Improve Application Quality on IEEE 802.11 Networks Choong-Soo Lee Ph.D. Dissertation Defense Worcester Polytechnic Institute June 10, 2010 Committee members: Mark Claypool (WPI) Robert Kinicki (WPI) Craig Wills (WPI) Carey Williamson (Univ. of Calgary)

2. Typical Home Network 2

3. What happens to application quality? 3 Quake IV: Moving around

4. What are the problems? Application quality degrades when multiple applications run at the same time. Average home users find it difficult to configure the network to maximize their experience. Applications change over time. Prioritizing an application with poor wireless connectivity degrades the performance of other applications. 4

5. What causes the quality degradation? Increased end-to-end delay ▫ A queue buildup along the downstream network path due to congestion ▫ Wireless losses and retransmissions Same treatment for all applications ▫ First-In-First-Out (FIFO) queue management Increasing bandwidth of home broadband access makes the wireless access point (AP) the bottleneck. 5

6. Current Approaches (1/2) Two-stage process: ▫ Classification ▫ Treatment Classification ▫ Applications can claim quality requirements. ▫ Middle boxes can figure out quality requirements. 6

7. Current Approaches (2/2) Treatment ▫ First-In-First-Out (FIFO)  No prioritization  The baseline performance ▫ Strict Priority Queue (SPQ)  A pre-determined, static set of priorities based on the classification  The best performance in most cases. 7

8. Requirements for “Smart” Access Point (AP) Prioritization Minimal configuration Application independence ▫ No changes to the end hosts ▫ No need to know specifics of applications ▫ No need to worry about emerging applications Ability to adjust priorities to dynamic wireless conditions ▫ Even one node with poor wireless connectivity can degrade the wireless network performance. 8

9. Summary of Current Approaches 9 RequirementsFIFOSPQ Prioritization ×√ Minimal configurationN/A × Application independenceN/A × Dynamic wireless conditionsN/A ×

10. Outline Introduction √ CHAP Approach CHAP Simulation CHAP Implementation Conclusion, Contributions and Future Work 10

11. Summary of Network Traffic Characteristics LowMediumHigh Low Medium High 11

12. Credit-based Scheduling Use of credit as priority Map application traffic characteristics to credit 12 Bandwidth Delay Tolerance Credit

13. Credit-based Home Access Point (CHAP) 13 Credits in time Wireless Transmission Time P

14. How CHAP Works 14 Credits S N1N1 N2N2 Delay sensitive Delay insensitive Wireless Access Point The delay insensitive flow runs out of credits.  Increase credits of all active flows by: 1. halving the current credits 2. increasing by boost parameter (15 bars) Cost of each transmission = 5 bars

15. How CHAP Works 15 S N1N1 N2N2 Wireless Access Point Cost of each transmission to N 1 = 5 bars Cost of each transmission to N 2 = 5 bars Credits Delay sensitive Delay insensitive 10

16. AP Requirements and CHAP 16 RequirementsCHAPNotes Prioritization √ Credits differentiate priorities of flows. Minimal configuration √ Boost parameter is the only additional configurable parameter to FIFO. Application independence √ CHAP does not require changes at end hosts. CHAP does not use anything specific (signatures, heuristics, etc.) from applications. Dynamic wireless conditions √ Credits are adjusted by frame transmission times.

17. CHAP Research Components Modeled CHAP analytically Simulated CHAP in an IEEE 802.11g infrastructure network against FIFO and SPQ using Network Simulator (NS2) Implemented CHAP as a queueing discipline in Linux ▫ Emulated 802.11g network over 100 Mbps Ethernet 17

18. Outline 18 Introduction √ CHAP Approach √ CHAP Simulation CHAP Implementation Conclusion, Contributions and Future Work

19. Simulation Overview Validation ▫ Simulation Validation ▫ Model Validation Performance evaluation varying: ▫ One application at a time: Game, VoIP, Video, Web ▫ Multiple applications: Game, VoIP, Video, Web, FTP ▫ Distance(s) of wireless nodes: 1 ~ 30 m ▫ Queue size: 35 ~ 350 packets ▫ Background application: FTP, P2P ▫ Network latency between end hosts: 2 ~ 150 ms ▫ TCP variants: NewReno, BIC, CUBIC, Compound 19

20. Settings: One Application at a Time 20 3090150210270330 Time (s) App. 2 App. 1 Queue Limit35 packets D 1, D 2 1 m L 1, L 2 1 ms Application 2FTP Queueing DisciplineFIFO, CHAP, SPQ Application 1Game, VoIP, Video, Web

21. Results: One Application at a Time 21 CHAP provides the quality of a delay sensitive application higher than FIFO and similar to SPQ, while maintaining the FTP throughput. +24%0%+65%+60% FIFOCHAPSPQ

22. Settings: Multiple Applications 22 Queue Limit35 packets D 1, D 2, D 3, D 4, D 5 1 m L 1, L 2, L 3, L 4, L 5 1 ms App. 1, App. 2, App. 3, App. 4, App. 5 Game, VoIP, Video, Web, FTP Queueing DisciplineFIFO, CHAP, SPQ 3090150210270330 Time (s) App. 5 App. 1 App. 2 App. 3 App. 4

23. Results: Multiple Applications 23 CHAP provides the quality of delay sensitive applications higher than FIFO and similar to SPQ, even when multiple applications run concurrently. +20%+3%+93%+51% FIFOCHAPSPQ

24. Settings: Varying Wireless Conditions 24 3090150210270330 Time (s) App. 2 App. 1 Queue Limit35 packets D2D2 1 m L 1, L 2 1 ms Application 1Video Application 2FTP Queueing DisciplineFIFO, CHAP, SPQ D1D1 1, 5, 10, 15, 20, 25, 30 m

25. Results: Varying Wireless Conditions 25 CHAP provides higher throughput than FIFO and SPQ CHAP provides the video application quality higher than FIFO and close to SPQ, while providing higher FTP throughput than FIFO and SPQ when the video application is placed further away from the AP.

26. 26 Outline Introduction √ CHAP Approach √ CHAP Simulation √ CHAP Implementation Conclusion, Contributions and Future Work

27. Implementation Overview Performance evaluation with actual user activities Implementation Validation ▫ without frame errors ▫ with frame errors Case Studies in a Residential Network ▫ Web Browsing ▫ Online Game 27

28. CHAP Implementation Hardware: IBM ThinkPad T60 ▫ CPU: Core2Duo T5500 (1.66GHz) (64-bit) ▫ RAM: 2 GB Software: Debian Linux (Lenny) ▫ Kernel version 2.6.26 ▫ Traffic Control (TC) + Queue Discipline (Qdisc) ▫ Emulated 802.11g over 100Mbps Ethernet 28

29. IEEE 802.11g Emulation Difficulties with the hostAP and madWifi driver ▫ Kernel panics and wireless node disassociation Emulated IEEE 802.11g over 100 Mbps Ethernet ▫ Delay packets on dequeue()  DIFS + SIFS + DATA and ACK transmission times ▫ Use of kernel timer interrupts  4096Hz kernel for the finest granularity (≈ 244 μs) 29

30. Case Study Setup Broadband: Verizon 7Mbps/768Kbps DSL Line The Linux bridge acts as the bottleneck. ▫ Extra 2ms delay  about 5 Mbps 350 packet queue limit Boost parameter of 25 ms for CHAP 30 Case Studies Web Browsing: Firefox 3.6 Online Game: Quake IV Moving around Shooting a shotgun

31. Websites Tested Yahoo Google News CNN Wall Street Journal 31

32. Webpage Load Times 32 CHAP provides similar webpage load times to FIFO in the absence of the file download, and provides much faster webpage load times than FIFO in the presence of the file download.

33. Quake IV Servers 33 Worcester, MA

34. Quake IV Server Ping Times 34 425 ms Playable Threshold CHAP provides similar server ping times to FIFO in the absence of the file download, and provides much smaller server ping times than FIFO in the presence of the file download.

35. Quake IV – Moving Around FIFO – Quake IV + Debian ISO Download (477 – 546 ms) CHAP – Quake IV + Debian ISO Download (62 – 94 ms) FIFO – Quake IV alone (62 – 125 ms) 35 CHAP allows the player to move around freely as if there is no concurrent file download while FIFO prevents the player from moving due to the large roundtrip time to the server.

36. 36 Outline Introduction √ CHAP Approach √ CHAP Simulation √ CHAP Implementation √ Conclusion, Contributions and Future Work

37. Conclusion (1/2) Credit-based Home Access Point (CHAP) improves quality of delay sensitive applications when multiple applications run concurrently. ▫ Quality higher than FIFO and close to SPQ CHAP requires only one additional configuration parameter (boost parameter), and it is preset to a recommended value of 25 ms. 37

38. Conclusion (2/2) CHAP does not rely on specifics of any applications, thus able to adapt to changes in applications over time. CHAP adjusts applications’ priorities according to dynamic wireless conditions and assists applications with good wireless connectivity. 38

39. Contributions Design and configuration of CHAP Analytical modeling of CHAP Performance evaluation of CHAP in NS2 Implementation and evaluation of CHAP Validation of NS2 simulation Design and implementation of a video streaming application in NS2 Analysis of TCP variants in NS2 39

40. Future Work Investigation of the boost parameter CHAP Implementation in a real AP ▫ DD-WRT ( http://www.dd-wrt.org/ ) Multiple APs in a home network Impact of upstream network traffic Application of CHAP in various types of wireless networks ▫ academic and corporate networks ▫ cellular networks Enhancement of the video application in NS2 40

41. Questions and Comments ? 41