1. Achieving Inter-Session Fairness for Layered Video Multicast Wei Kyang Lau and Chieh Ying Pan, Department of computer science and engineering, National Sun Yat-Sen University

2. Introduction To handle heterogeneous network conditions, layered multicast algorithms are used to solve this problem Receiver-driven layered multicast (RLM) is proposed in 1996 Join-experiment Decide to add a layer or drop a layer every join-timer (Tj) Back-off action If the join-experiment results in congestion, it doubles the join-timer (Tj) for next join-experiment Tj would be decreased if no packets loss for a period of time

3. Introduction (Cont.) Drawbacks They do not provide fair bandwidth sharing between competing video streams. Receivers can ’ t respond to the congestion quickly Two schemes without layering (1998) Priority Dropping Different priorities are assigned to the I, B and P frames in the MPEG video stream. More important frames such I frames are assigned the higher priority. In congestion, the frames with the lowest priority are dropped first Uniform dropping

4. Introduction (Cont.) Comparison between RLM and priority dropping Priority dropping does not have the problem of fairness Drawbacks of priority dropping Implementing priority dropping at routers is more complex Providing no incentives for receivers to lower their subscription level

5. Introduction (Cont.) Layer-based congestion sensitivity (1999) Different layers have different sensitivity to congestion r: the loss rate threshold for congestion t: the congestion resilience time Sensitive functions: f r () and f t ()

6. Introduction (Cont.) Layer-based congestion sensitivity(cont.) The main idea is to adjust r and t to make higher layer drop easier

7. Introduction (Cont.) Two schemes are proposed to achieve fairness: Layered video multicast with congestion sensitivity and adaptive join-timer (LVMCA) Layered video multicast with priority dropping (LVMPD)

8. Layered video multicast with congestion sensitivity and adaptive join-timer (LVMCA) LVMCA To reduce convergence time (the time needed for all sessions to come to their steady reception levels) and the variations of video transmission rates after convergence Modify the join-timer of layer-based congestion sensitivity Allow low-level video session to add a layer easier Decrease the time to converge to fairness Make adding high-level layer harder, decrease unnecessary congestion caused by the join-experiment of high-level sessions while there is no spare bandwidth.

9. LVMCA (Cont.) Define a join-timer function T j = f j (), where f j (i) = J+1, J is a constant.

10. Layered video multicast with priority dropping (LVMPD) In LVMPD, it adapt the idea of sending video packets over various layers and combine it with priority dropping to improve video quality and achieve better inter-section fairness Assign higher priority to lower layers and lower priority to high layers.

11. LVMPD (Cont.)

12. Simulation results Simulation topology

13. Simulation results (Cont.)

14. 104 sec

15. Simulation results (Cont.) LVMCA 90 sec

16. Simulation results (Cont.) LVMPD 67 sec

17. Conclusion LVMCA could shorten the convergence time to fairness and leads to steadiness LVMPD could converge to fairness with the time even shorter than the LVMCA and rates of vide streams are more steady after convergence