Transporting Compressed Video Over ATM Networks with Explicit-Rate Feedback Control

1. Transporting Compressed Video Over ATM Networks with Explicit-Rate Feedback Control IEEE/ACM Transactions on Networking, VOL. 7, No. 5, Oct 1999 T. V. Lakshman, Senior Member, IEEE, Partho .P. Mishra, and K. K. Ramakrishnan, Associate Member, IEEE

2. Abstract • A scheme for transmission of VBR compressed video for interactive applications using the explicit-rate congestion-control mechanisms is proposed.

3. Mechanisms for transporting video traffic • CBR Transport • The output of a video coder is locally buffered at the coder to convert it into a CBR stream. • Admission control is simple. • There is no attempt to exploit any multiplexing gains possible in the original VBR traffic. • VBR Transport • The traffic generated by the coder is transported in a completely unrestricted manner. • The effective bandwidth needed is less than that for CBR video of the same quality. • The source model is loss of efficiency.

4. Mechanisms for transporting video traffic (Cont’d) • Renegotiated CBR (RCBR) • A video coder generates a piecewise linear CBR stream with periodic renegotiation of the bit rate between the coder and the network. • RCBR may be viewed as a hybrid of CBR and VBR. • The source is unable to make use of the newly available bandwidth until the next renegotiation instant. • UBR, Best-Effort Transport • Video sources continuously estimate the available bandwidth and adapt to it. • Quality can get unacceptably poor, since there is no minimum rate guaranteed.

5. Definition of rates • Nominal Rate • The rate that is required by the encoder to code the frame at ideal quality. • Target Rate • The rate given to the encoder based on the algorithm for smoothing and rate-adaptation. • Demand Rate • The rate that the source request from the network. • Allowed Rate • The rate returned from the network.

6. Operation of the feedback-control mechanism • A source specifies a demand rate in each transmitted RM cell in the ER field. • Switches compute the rate they may allocate to each VC, and overwrite the ER field with the computed allocated if it is lower than what was in the ER field of the received RM cell. • On reaching its destination, the RM cell is returned to the source, where now sets its transmit rate based on the allocated in the ER field of the returned RM cell.

7. Operation of the feedback-control mechanism (Cont’d) • When an RM cell returns with an allocated rate ER, the source’s allowed rate is changed as follows: if ACR  ER ACR = max(min(ER, DEMAND), MCR) else ACR = max(min{ACR+RIF*PCR), ER}, MCR) • When an RM cell is transmitted, the ER field is set to max(DEMAND, ACR). RM cells are periodically transmitted.

8. Enhancement: Weighted max-min fairness • The goals • Satisfied VC’s should receive an allocation equal to their request ER. • The amount of extra capacity left over from the allocation to satisfied VC’s is shared among bottlenecked VC’s in proportion to their demands Di. • To achieve a weighted max-min fair allocation, we require the original demand from the source to be available to all of the resources in the network.

9. Enhancement: Weighted max-min fairness (Cont’d) • An additional field, called “source demand”, in the RM cell is introduced for this purpose. • The weight at a bottleneck for a source i whose demand is Di is given by Where F is the set of flows placing a demand on this resource.

10. Simulation results

11. Simulation results (Cont’d) 5.88/4.94 = 1.190 5.71/4.80 = 1.189 6.11/5.12 = 1.193 6.31/5.28 = 1.195 … 1.74/1.39 = 1.252 1.32/1.19 = 1.109 1.77/1.46 = 1.212 1.78/1.59 = 1.119 …

12. Simulation results (Cont’d) As the number of active connections is increased, the mean and variance of the end-to-end delay increases.

13. Simulation results (Cont’d)

14. Simulation results (Cont’d)