1. Advanced Technology Laboratories Practical Considerations for Smoothing Multimedia Traffic over Packet- Switched Networks Christos Tryfonas tryfonas@sprintlabs.com April 2002 Sprint Advanced Technology Laboratories Burlingame, California

2. Advanced Technology Laboratories Problem Statement  Objective: Compute a rate schedule to send a video stream that maximizes some network parameter (e.g. utilization)  Constraints: Avoid decoder buffer under/overflows Network Interface Packet-Switched Network Video Server Stream Decoder (Buffer) Sender Receiver

3. Advanced Technology Laboratories Work-Ahead Smoothing Algorithms  Most smoothing algorithms compute maximum constant rate segments  Differences in the selection of the starting point of next segment Size of frame 4 Decoder Buffer Size Underflow Envelope Overflow Envelope Time Received Data Schedule Jitter

4. Advanced Technology Laboratories Outline  Practical Observations  Smoothing under variable network delays  Sender-Based Smoothing  Clock-Aware Smoothing  Renegotiation Failure Decisions  Conclusions

5. Advanced Technology Laboratories Practical Observations  Network delay is variable  Current smoothing algorithms compute schedule based on the receiver side  Not all data sent by the sender is utilized by the receiver  Some applications require clock to be recovered at the receiver (e.g. Broadcast TV)  Consideration of Renegotiation Failures

6. Advanced Technology Laboratories Smoothing under Variable Network Delays  Jitter usually a function of the reserved rate (e.g. Fair Queueing schedulers)  Currently optimization done with a pre-specified worst- case network jitter, i.e., not optimal  Overflow envelope is a function of the decision, not fixed  The burstiness of the source at the selected rate defines the ambiguity zones  

7. Advanced Technology Laboratories Our Approach  Optimization should be done by considering the ambiguity zones resulting from the varying reserved rates  Existing optimization algorithms should consider the variable ambiguity zones for their decisions Size of frame 4 Decoder Buffer Size Underflow Envelope Overflow Envelope Time Received Data Schedule – Ambiguity Zones

8. Advanced Technology Laboratories The Case of Rate Decrease  New rate corresponds to higher jitter  Lower envelope is changed to accommodate the increase in jitter Schedule Ambiguity Envelope Time Received Data Jitter 2 Jitter 1 Jitter 1 < Jitter 2

9. Advanced Technology Laboratories The Case of Rate Increase  New rate corresponds to less jitter  A peak-rate segment is inserted before the new ambiguity zone with the new jitter Schedule Ambiguity Envelope Time Received Data Jitter 2 Jitter 1 Jitter 1 > Jitter 2 Peak Rate segment

10. Advanced Technology Laboratories  Two approaches: 1.Rate Adaptation 2.Forward Schedule computation Sender-Based Smoothing  Not all data transmitted is inserted in the decoder buffer Sequence Header GOP Header Frame Header Frame Coding Extension Frame Header Frame Coding Extension Frame Data Actual Received Data after Rate Adaptation Time Received Data Received Data function as computed by optimization

11. Advanced Technology Laboratories Clock-Aware Smoothing Time Received Data Schedule Time Received Data Clock-Aware Schedule Clock Time Clock Time

12. Advanced Technology Laboratories Renegotiation Failure Decisions  Dynamic Requantization  Use of scalable-encoded streams  New Approach: Pre-computed alternate schedules with convergence points Time Received Data Main Schedule Alternate Schedule Renegotiation Failure Intersection Point Convergence Point

13. Advanced Technology Laboratories Conclusions  Current smoothing algorithms produce infeasible schedules in practice  Optimization has to consider: Variable Network Jitter a function of the reserved rate Unused data sent by the sender but not utilized by the receiver/decoder Clock-awareness Renegotiation Failures  For more  www.sprintlabs.com/People/tryfonas