1. A3: APPLICATION AWARE ACCELERATION FOR WIRELESS DATA NETWORKS Athours: Zhenyun Zhuang and Tae-Young Chang GNAN Research Group, Georgia Tech, Atlanta, GA Raghupathy Sivakumar and Aravind Velayutham Asankya Networks, Inc., Atlanta, GA MobiCom’06

2. OUTLINE Motivation Methodology Transaction Prediction (TP) Redundant and Aggressive Retransmissions (RAR) Prioritized Fetching (PF) Infinite Buffering (IB) Application-aware Encoding (AE) Integrated A3 Results Pros and Cons

3. MOTIVATION Wireless Environments High loss rate Large delay Low bandwidth Improving the performance of popular applications in wireless environments Common Internet File Sharing protocol (CIFS) Simple Mail Transfer Protocol (SMTP) Hyper-Text Transfer Protocol (HTTP) Figure from Business Communications Review (April 2006) Common Internet File Sharing protocol(CIFS)

4. A3: APPLICATION-AWARE ACCELERATION Application aware Recognize applications Application transparent No modifications to applications A set of design elements Transaction Prediction (TP) Redundant and Aggressive Retransmissions(RAR) Prioritized Fetching (PF) Infinite Buffering (IB) Application-aware Encoding (AE)

5. A3: APPLICATION-AWARE ACCELERATION Implementation with Netfilter for Linux Systems Using Netfilter hooks to capture the packets from traffic Deployment Model

6. EXPERIMENTAL SETUP Application Emulator (AppEm) Using IxChariot to generate accurate application specific traffic patterns A3 Emulator (A3Em) Wireless Network Emulator(WNetEm) are implemented with the framework of NS2

7. TRANSACTION PREDICTION (TP) - DESIGN CIFS requests only 16 – 32KB data in a request When requested data < BDP Low utilization Not pre-fetching Deterministically predict future requests Issue requests ahead of time CIFS Traffic Patterns

8. TRANSACTION PREDICTION (TP) - DESIGN (CONT.) CIFS Traffic Patterns

9. TRANSACTION PREDICTION (TP) - ISOLATED RESULT

10. Explicit Loss Notification(ELN) Only reduce the congestion window due to congestion loss. Wireless TCP(WTCP) Any lost segments will have to wait in the WTCP's buffer until the first one is confirmed to have been received.

11. TRANSACTION PREDICTION (TP) - ISOLATED RESULT

12. Redundant and Aggressive Retransmissions(RAR) - Design RAR helps to protect thin session control messages from losses Not applying to DATA SMTP Traffic Patterns

13. The interval between each redundant packets RTTavg/10 RTTavg is the average RTT observed so far (dynamical update) Aggressive retransmission Setting a timer = RTTavg + c c is a small guard constant If client gets response stop the timer If all copies of a message are lost Time out=>Retransmission Redundant and Aggressive Retransmissions(RAR) - Design

14. REDUNDANT AND AGGRESSIVE RETRANSMISSIONS(RAR) - ISOLATED RESULT

15. Prioritized Fetching (PF) - Design Using a plugin of web browser to query current focus window and scrolling information The visible object will be downloaded immediately If the object is not visible, then download it later

16. PRIORITIZED FETCHING (PF) - ISOLATED RESULT

17. INFINITE BUFFERING (IB) - DESIGN Mobile device may have small connection buffer and reading slowly from connection buffer Setting advertised window to MAXIMUM value 2KB data ACK window=0 ACK window=2K 2KB 0KB 2KB

18. INFINITE BUFFERING (IB) - DESIGN Mobile device may have small connection buffer and reading slowly from connection buffer Setting advertised window to MAXIMUM value Exceeded data will be stored to 2 nd storage 2KB data ACK window= MAXIMUM 2KB 0KB 2KB 2KB data 2 nd Storage

19. INFINITE BUFFERING (IB) - ISOLATED RESULT How fast application layer reads the data? What is the size of connection buffer?

20. Application-aware Encoding (AE) - Design Uses application and user specific information SMTP – Using previous mails Only available to a pair user Not binary compression(such as ZIP) 10 Persons (100 emails)

21. Application-aware Encoding (AE) - Design A > B Hi B, Do you have come to play a game with us? B > A Hi A, Sorry, I can’t go to play a game with you. => 0x1 A, Sorry, I can’t go 0x2 0x3 a 0x4 0x5 you Using 10-16bit to represent a word(6bytes)

22. APPLICATION-AWARE ENCODING (AE) - ISOLATED RESULT

23. INTEGRATED A3 RESULTS

24. PROS This paper presents a way to improve the performance of the applications obviously with application transparent Did not present that how the processing overhead affects the performance Infinite Buffering result CONS

25. TRANSACTION PREDICTION (TP) - IMPLEMENTATION

26. REDUNDANT AND AGGRESSIVE RETRANSMISSIONS(RAR)

27. PRIORITIZED FETCHING (PF)

28. Infinite Buffering (IB)

29. Application-aware Encoding (AE)