1

Cooperative Inter-stream Rate Control Scheme for Layered

Multicast

Masato KAWADA, Hiroyuki MORIKAWA , Tomonori AOYAMA,

School of Engineering,

The University of Tokyo

2

Introduction

• Sharing streams– If streams pass the same bottleneck link and

share the limited bandwidth, we call it sharing streams in the following

• Design a end-to-end inter-stream control scheme for layered multicast which enables to adjust the allocated bandwidth of sharing streams according to the predefined policy

3

Introduction

• Use Receiver-driven layered Multicast (RLM) as basic method for receiving layered multicast streams

• Establish Multiple Streams Controller(MSC) mechanism over RLM at each receiver.

4

Receiver-driven Layered Multicast (RLM)

• Receiver-driven rate control has tow main mechanism– The probe of empty bandwidth with trial

joining to the next layered and loss detection– Rate degradation on congestion by dropping the

extended layer to reduce the congestion

• Join-experiments– A receiver subscribes to an additional layer and

observes if the subscription degrades its reception quality as opposed to improving it

5

Receiver-driven Layered Multicast (RLM)

• Detection time– The time due to the experiment in a certain time

• Join-timer– The interval to the next join-experiment

• Subscription Success– No congestion occurs and packet loss rate does

n’t exceed a configured threshold during the detection time

6

Receiver-driven Layered Multicast (RLM)

• Subscription Fail– Congestion occurs due to the experiment during

the detection– Packet loss rate measured in a detection time

exceeds a configured threshold

• A receiver multicasts a join-experiment report identifying the experimental layer to other receivers when doing the experiment

7

Receiver-driven Layered Multicast (RLM)

• Disadvantages– Receivers can’t adjust the allocated

subscription level among streams

– If there are streams occupying the bottleneck link, the new stream passing the same link has little available bandwidth

8

Receiver-driven Layered Multicast (RLM)

• Disadvantages(Cont.)– If join-experiments are overlapped among

streams, the experiments are likely to interfere with each other and be failed

– If temporary congestions caused by join-experiments of other streams happen continually, a receiver misrepresents this congestion as continuous one and drops the highest layer

9

Overview of Inter-Stream Control

RLM++ with MSCRLM

10

Overview of Inter-Stream Control

11

RLM++ Operation Support

Tj

12

RLM++ Operation Support

• MSC can prevent the wasteful decreasing of the subscription level happening by misrepresenting congestion due the other stream’s join-experiments as continuous congestion

13

Construction of Congestion Table

14

Construction of Congestion Table

• MSC grasps the sharing state of bandwidth in a bottleneck link both among receivers and among streams from congestion table.

• The information in the table becomes the clue of performing the adjustment of subscription level among streams

15

Base Layer Protection

• MSC and RLM++ should guarantee the bandwidth for the base layer by degrading the subscription level of the other stream

• MSC multicasts the dropping request and the receivers who detect congestion caused by the base layer decrease the subscription level when they receive the request

16

Adjustment of Subscription Level

• MSC can find the situation of dropping or subscription among streams by analyzing congestion table

• MSC can adjust the subscription level among streams according to receiver;s request or a predefined rule

17

Simulation and Evaluation

4 Multicast Layer: 100kbps, 100kbps,

200kbps, 400kbps

Request sequence:

Bottleneck:

Topology (1): 5 — 6 (300 kbps)

Topology (2): 4 — 5 (600 kbps)

Topology:

Simulation Parameters:

18

Results and Evaluation

• Base layer protection:– R2 joins Stream ‘z’ at 100 sec.

Base layer join (no protection) Base layer join (protection)

Y

X

Z

19

Results and Evaluation

• Adjustment of subscription level– R1 changes the subscription level from (x, y) =

(2, 1) to (x, y) = (1, 2)

20

Results and Evaluation

• Scalability– The three senders in topology (2) starts their

own session randomly during the first 5sec of simulation time, and receivers start receiving each streams at random time between 5sec and 7sec

– Three metrics• Convergence time• Loss rate• Subscription level stability

21

Results and Evaluation• Scalability --- Convergence time

– The time between when it starts and when it reaches its stable subscription level

22

Results and Evaluation

• Scalability --- Loss rate

23

Results and Evaluation

• Scalability --- Stability of subscription level– We use the frequency of missdropping layers as

evaluation values of subscription level stability

24

Conclusion

• MSC and RLM++ can do the base layer protection and the adjustment of subscription level among streams by constructing and analyzing congestion table

• Effective bandwidth allocation among streams with user’s request and receiving streams with stable quality are possible