18
1 Multi-layer Active Queue Management and Congestion Control for Scalable Video Streaming Kang, S.-R.; Zhang, Y.; Dai, M.; Loguinov, D.; Distributed Computing Systems, 2004. Proc eedings. 24th International Conference on , 24-26 March 2004

Multi-layer Active Queue Management and Congestion Control for Scalable Video Streaming

  • Upload
    miya

  • View
    48

  • Download
    0

Embed Size (px)

DESCRIPTION

Multi-layer Active Queue Management and Congestion Control for Scalable Video Streaming. Kang, S.-R.; Zhang, Y.; Dai, M.; Loguinov, D.; Distributed Computing Systems, 2004. Proceedings. 24th International Conference on , 24-26 March 2004 . Core components. Priority Active Queue Management - PowerPoint PPT Presentation

Citation preview

Page 1: Multi-layer Active Queue Management and Congestion Control for Scalable Video Streaming

1

Multi-layer Active Queue Management and Congestion Control for Scalable Video Streaming

Kang, S.-R.; Zhang, Y.; Dai, M.; Loguinov, D.;Distributed Computing Systems, 2004. Proceedings. 24th International Conference on , 24-26 March 2004

Page 2: Multi-layer Active Queue Management and Congestion Control for Scalable Video Streaming

2

Core components

Priority Active Queue Management mark packets of different importance and drop

less important packets first Congestion Control

feedback network information from router and adjust the frame size

Partitioned Enhancement Layer Streaming (PELS) priority marking, AQM, congestion control

Page 3: Multi-layer Active Queue Management and Congestion Control for Scalable Video Streaming

3

Outline

Background Best-effort network is not enough AQM Congestion control Simulation Conclusion

Page 4: Multi-layer Active Queue Management and Congestion Control for Scalable Video Streaming

4

Goal

majority of packets across bottleneck carry useful information

retransmission-free

Page 5: Multi-layer Active Queue Management and Congestion Control for Scalable Video Streaming

5

MPEG-4 FGS

base layer is more important than enhancement layer

Page 6: Multi-layer Active Queue Management and Congestion Control for Scalable Video Streaming

6

Best-Effort Streaming

assume independent Bernoulli packet loss with probability p, expected number of useful packets (consecutively received) is

Page 7: Multi-layer Active Queue Management and Congestion Control for Scalable Video Streaming

7

Best-Effort Streaming

Page 8: Multi-layer Active Queue Management and Congestion Control for Scalable Video Streaming

8

Optimal Preferential Streaming goal: achieve U = 1 in order to be optimal, upper layer should be

dropped before lower layer enhancement layers further divide into two

layers

Page 9: Multi-layer Active Queue Management and Congestion Control for Scalable Video Streaming

9

Active Queue Management

two types of queues: PELS queue and Internet queue

Weighted round-robin (WRR)

Page 10: Multi-layer Active Queue Management and Congestion Control for Scalable Video Streaming

10

Active Queue Management

send low-priority packets only after all high-priority packets are sent

no end-user can gain by marking all packets with highest priority

Page 11: Multi-layer Active Queue Management and Congestion Control for Scalable Video Streaming

11

Selection of γ pR = pxi/γxi = p/γ = pthr

optimistic: pthr~1 U~1 pessimistic: pthr~p γ =1

yellow layer = (1- γ)xi = 0 close-form expression for γ

Page 12: Multi-layer Active Queue Management and Congestion Control for Scalable Video Streaming

12

Selection of γ

when p=0.1, pthr=0.75, U>=0.96 when p=0.01, pthr=0.75, U>=0.996

Page 13: Multi-layer Active Queue Management and Congestion Control for Scalable Video Streaming

13

Congestion control

modified from Kelly’s control (a game-theoretic and optimization method), discrete version called Max-min Kelly Control (MKC)

reduce bitrate and keep waste to minimum

Page 14: Multi-layer Active Queue Management and Congestion Control for Scalable Video Streaming

14

Simulation

ns2 simple bar-bell topology 1 video frame

= 63000 bytes = 126 packets(21 base layer packets)

50% bottleneck forTCP cross traffic

Page 15: Multi-layer Active Queue Management and Congestion Control for Scalable Video Streaming

15

Stability properties ofγ show that

γis stable in the close-form expression with dynamic loss prob;

by using found γ , loss prob of red packets kept to target threshold

Page 16: Multi-layer Active Queue Management and Congestion Control for Scalable Video Streaming

16

Delay characteristic of PELS

Page 17: Multi-layer Active Queue Management and Congestion Control for Scalable Video Streaming

17

Fairness of MKC congestion control

Page 18: Multi-layer Active Queue Management and Congestion Control for Scalable Video Streaming

18

Conclusion

preferential streaming framework (PELS) provides high level of end-user QoS

independent of underneath congestion control methods