1. 1 Real-Time Traffic over the IEEE 802.11 Medium Access Control Layer Tian He J. Sobrinho and A. krishnakumar

2. 2 Outline Motivations Possible approaches A proposed solution in the paper Evaluations Conclusions & Comments

3. 3 Motivation Real time applications are ever more popular – VOIP Market $5b by 2005. – Myriad of streaming services: VOD, Video Phone, D-Sharing. Videoconferencing. Major service under wireless network. – Currently data is dominate service in wired network, while Data is “special service” in wireless network. real-time streaming is dominate market in wireless environment IP networking towards wireless, mobile environment. Inherently it is a Interesting research problem

4. 4 Required QoS Real time traffic is not too sensitive to delay – ~400ms for VOIP, ~250ms for video conferencing Very sensitive to jitter – As little as 150ms can be unpleasant. – VOIP specification require average e2e delay 145ms. Effect of lost packets is strongly codec dependant.

5. 5 Even harder in wireless Narrow bandwidth available – 802.11a:54Mbps 802.11b:11Mbps 802.11g/e:22Mbps – IEEE 802.3ae : 10Gbps 909 times faster High control overhead – large synchronization fields – larger MAC headers 34B vs 14B in 802.3 – more management packets (AP registration) Inherent contention media (open space)

6. 6 Why not 802.11 DCF A wants to transmit but channel is busy B A C Packet to Node C RTS CTS Contention slots ACK positive acknowledgment

7. 7 Existing solution 802.11 centralized approach: PCF to guaranteed QoS.

8. 8 Why not 802.11 PCF Centralized PCF Scheme – Single point of failure – Single media, no space multiplexing – High overhead. (Registration, Polling ) – In-compatible with multi-cell setting.

9. 9 Other Solutions for Real-time Traffic Time Division Multiple Access (TDMA) – Fixed slotting: Inefficient – dynamic slotting: complex scheduling algorithm Code Division Multiple Access (CDMA) – Fixed coding length: inefficient – Dynamic coding: dynamic code assignment Token Ring Passing – Only suitable for single contention media

10. 10 Key ideas in this paper DCF mode for data stations. Special mode for real-time stations. Real-time stations have priority over data station by using shorter IFS. Real-time stations proactively send “black bursts”, of length proportional to waiting time. Guarantee one and only one real-time station wins for each contention phase.

11. 11 Assumption Every node can sense each other’s transmissions (no hidden/exposed terminal problem). No RTS/CTS is used. Real-time stations periodically send out packets at same rate.

12. 12 Definitions

13. 13 Access procedures 1. Single data station access procedure 2. Interactions among data stations 3. Single real-time station access procedure 4. Interaction among real-time stations 5. Interaction between data stations and real- time stations

14. 14 1. Single data station access CSMA/CA as access procedure. Contention Window Backoff-Window Busy Medium T long T short T long T short DATAACK DATA t A A

15. 15 Difference from 802.11 Standard Data stations keep sensing the channel even no packets ready for transmission. 802.11 only senses the channel when need. It use the pervious channel status to decide whether back-off or transmit immediately.802.11 needs DIFS delay before make a decision. This scheme consumes more energy, but has shorter delay.

16. 16 2. Interactions among data stations CSMA/CA as access procedure. Contention Window Backoff-Window Busy Medium T long T short DATA t Contention Window Backoff-Window Busy Medium T long T short A B

17. 17 3. Single real-time station access Contention Window T long Busy Medium T long T short T med DATA T short DATA ACK T obs A A

18. 18 4. Interaction among real-time stations Round robin access among real-time stations Busy T med T obs Busy T obs Data Busy A B T med Busy T obs Data Busy

19. 19 4.Interaction among real-time stations How to set T obs. – T obs must be shorter than a black burst slot, otherwise we station A will not back off. – T obs must be shorter than T med, so that no real-time station will access channel during observation. T med T long Busy Medium T obs T med T long Busy Medium T obs Data Schedule Time A B Data

20. 20 5.Interaction between real-time and data stations

21. 21 Negative Acknowledgement Positive acknowledgement has an efficiency penalty. When receiver gets a packet from sender at time T, it expects another packet at time: T+ t sch +t obs. When receiver do not receive the packet at expected time interval, it sends out a negative acknowledgement.

22. 22 Theoretical Analysis: Stability Definition 1: – The system is stable if and only if whatever the initial conditions is, there is an L >=0, such that the access delay for real-time station is zero after L rounds (converge) Definition 2: – The system is unconditionally stable if and only if it is stable no matter the magnitude of the perturbation T (Overshoot independent)

23. 23 Stability (contd) DATA Delay

24. 24 Stability conditions The system is unconditionally stable if and only if following inequality holds N is #real-time station (1) In addition if, the system is stable if and only if following inequality holds T is initial disturbance (2)

25. 25 Nominal values

26. 26 Simulation Result (1)

27. 27 Simulation Result (2)

28. 28 Simulation Result (3)

29. 29 Conclusions Distributed access Higher priority for real-time station to access the channel Can be overlaid on 802.11 without changing data stations Virtual TDMA structure for real-time stations  constant access rate. Under stable condition  bounded access delay

30. 30 Critical comments Possible data contention between real-time stations,even assume no hidden & exposed channel problem. RTS/CTS is desired. Fix channel access interval for real-time stations ( round robin), which is inefficient. Only consider initial disturbance T. No stable analysis for periodic or sporadic disturbance. Evaluations on how data stations impact the performance of real-time stations are more desired: converge(settling) time vs initial disturbance