V. Fodor and Gy. Dan, KTH - Marholmen 2002

End-to-end control for audio-visual communication

Viktoria Fodor and György DánLaboratory for Communication Networks,

KTH/IMIT{viktoria,gyuri}@it.kth.se

V. Fodor and Gy. Dan, KTH - Marholmen 2002

Outline

• Overview of audio-visual communication, requirements of QoS provisioning

• The end-to-end argument

• Source shaping in networks with small buffers

• quality differentiation with forward-error correction

• Future plans

V. Fodor and Gy. Dan, KTH - Marholmen 2002

Problem definition

• Audio-visual communication requires limited packet loss probability, end to end delay and jitter

• Audio-visual traffic can not be served as best effort

• How to provide QoS guarantees?– Considering the existing Internet architecture

– Following the end-to-end argument

V. Fodor and Gy. Dan, KTH - Marholmen 2002

The end-to-end argument

• Help to define new control functions for new applications (Saltzer, Reed and Clark, ‘81)

Control functions:• implemented fully inside the network• for all traffic• necessary for acceptable performance• future proof - does not limit future applications

Control functions:• can not be solved inside the network• application specific

V. Fodor and Gy. Dan, KTH - Marholmen 2002

Proposed solution

• Inside the network:– dedicated capacity for a limited set of service class, thus per class

scheduling at the routers• best effort

• controlled load - typically for audio-visual communication

• guaranteed service

– separate buffers for the service classes• controlled load: small buffers to limit delay and jitter

• At the network edge - for audio-visual communication:– probe based admission control to limit loss probability

(Prof. Gunnar Karlsson, Ingnacio Mas)

– traffic control functions to improve performance

V. Fodor and Gy. Dan, KTH - Marholmen 2002

Control functions

• Source shaping

• Forward-error correction

• Optimization problem:– there is a limited delay control functions can utilize

({end-to-end delay limit} - {network delay})

– performance of the control functions is proportional to the introduced delay

– how to allocate delay to the control functions?• performance of source shaping

• performance of forward-error correction with and without shaping

V. Fodor and Gy. Dan, KTH - Marholmen 2002

Source shaping

• To limit burstiness (peak rate) and thus increase the efficiency of buffering at the nodes

• Simplest shaper: leaky bucket

• Introduced delay: shaper buffer

(p,m) (c, p>c>m)B

V. Fodor and Gy. Dan, KTH - Marholmen 2002

Performance - source shaping

• Traffic mix (MPEG) with and without shaping

• Source shaping can decrease the average loss probability with many orders of magnitude with delay of 20-100ms

• Shaped flows have lower packet loss probability, ca. half of the loss of unshaped flows at the interesting loss levels

V. Fodor and Gy. Dan, KTH - Marholmen 2002

Performance - source shaping

• Shaped sources have more even loss distribution:– consecutive losses

– losses in block of packets

– losses in different frame types

• Shaping improves quality:– decreased average loss

– improved loss distribution

Relative loss (I) = loss in I / average loss

V. Fodor and Gy. Dan, KTH - Marholmen 2002

Forward-error correction

• Add error correction to a block of packets and regenerate lost packets at the receiver

• Introduced delay: ~ the block+code size

• Also increased network load ( increased loss probability)

information

error code

V. Fodor and Gy. Dan, KTH - Marholmen 2002

Performance - forward-error correction

• Traffic mix with different error correcting capability

• Error correction provides solution for quality differentiation(with the cost of higher source rate)

V. Fodor and Gy. Dan, KTH - Marholmen 2002

Conclusion

• Efficient audio-video transmission might be possible in networks with little control at the nodes

• Suggestions– capacity allocation for different service classes

– one class for delay and loss sensitive traffic

• end to end call admission control to limit loss

• use of small buffers inside the network to limit delay

• control mechanisms at the edge (shaping + forward error correction) to increase efficiency

V. Fodor and Gy. Dan, KTH - Marholmen 2002

Further work

• Forward-error correction and shaping– mathematical model to combine the two functions

• Adaptive forward-error correction– with fixed peak rate, changing the source coding rate (signal

processing)

– with feedback from the destination (control theory)