Network_coding_media..

On Network Coding Based Multirate Video Streaming in Dire

cted Networks

Chenguang Xu and Yinlong XuUniversity of Science and Technology of China

Outline

• Multirate Video streaming• About Network coding• Related works

– Without network coding– With network coding

• My work• Future work

Multirate Video Streaming • Property of Internet

– Heterogeneity of Receivers• Approaches:

– The replicated stream approach– Cumulative Layer approach, Such as MPEG

x (Layered Coding)– Non-cumulative Layer approach, Such as

MDC

An example of Layer Coding

What is Network Coding?

Transmission with network codingPacket level encoding at intermediate

nodesDecoding at receivers

The common method of network coding Linear Coding

E.g. 2a+3b

An example -Network Coding

T1 F1

T3

F3

F2

T2

S?

a

aa

b b

b

a+b

The advantages of Network Coding in Multicasting cont

Advantages: Throughput, Delay, Disadvantages: Packets Overhead(3%) Encoding/Decoding Time It

depends…

Related works-Streaming Without Network Coding

Layered multicast streaming without network coding Rate allocation for each layer

Fairness Issues Adaptive layer receiving

Related works-Streaming With Network Coding

Layered streaming with network coding “On multirate multicast streaming using network coding” A

llerton05 Encode the packets of different layers Objective: Maximize the total rates of receivers

Weakness: May cause ineffective transmission. Receiving higher layers while missing some lower layers .

My work-The Unachievabilityof Network Coding for Streaming

layer1

layer1

?k1*a+k2*b+k3*c+k4*d

{a, b, c, d}

m1*a+m2*b+m3*c+m4*d

Layer1 {a, c}Layer2 {b, d}2 time units as a generation

Conventional Content Distribution Streaming

k1*a+k2*b+k3*c+k4*d

m1*a+m2*b+m3*c+m4*d

1 2 3 4

1 2 3 4

* * * *

* * * *

k a k b k c k d

m a m b m c m d

For T1 and T2

Problem Descriptions

The Model1) Directed Networks G(V,E,c)2) a set of layers {Layer 1, Layer 2,…Layer k} ,

with a fixed rate rm for layer m3) R is the receivers’ set

Objective : Maximizing the total layers received

Basic Assumptions

• Each encoding generation occupying Δ consecutive time units.

• The buffer is large enough and the link state is stable.

• Acyclic network• Fixed rate for each layer

The Coding Scheme- LSNC

• Layered Separated Network Coding

• The Advantages of LSNC The advantage of network coding

Layer Separated for different priorities of layers Needn’t to pad the shorter packets with 0s

The Coding Scheme- LSNC cont

The remaining problems:1. How to determine the layer for each

receiver?2. How to allocate bandwidth for each layer?3. How to achieve the rate of each layer?

By existed network coding algorithm

Optimal Layer Separated Network Coding

OLSNC: Jointly Solve 1 and 2.( )

1 1

1 2 2 3, ( ) 1 ( )

( , ) ( , )

{ |( , )

, , (1)

0 1 ,1 ( ) (2)

0

i

i i

j i

OL vm

iji j

i i i i iOL v iOL v i

ij i i

l t l tij ji

j v v

Maximize

Subject to

for v R

or for v R j OL v

x x

0

{ |( , ) } }

( , )

{ |( , ) }

( , )0

{ |( , ) }

, 0, , 1 ( ) (3)

* , 1 ( ) (4)

,

i j

j i

j

j v v E E

t t

l iji il l i i

j v v E

l tj tl l t

j v v E

for i t i v R l OL v

x r for v R l OL v

x r for v R

( , )

1 ( ) (5)

( , ) , ,1 ( )

( , ) (6)

ij

t

l l tij i j t t

lij i j

l

l OL v

Y x for v v E v R l OL v

Y C v v

OLSNC-An exampleS is the source. T1 and T2 are receivers.The stream is consisted of 3 layers-L1, L2, L3, with rate of 1, 1, 1 respectively.

By OLSNC, T1 can get 2 layers, and T2 can get 3 layers.

{L1} {L1} {L2} {L2}

{L3}

OLSNC-An exampleWithout Network Coding:

Optimal Multicast Tree: T1 : 1 layerT2 : 2 layers

Optimal Multicast Sub-graph: T1 : 2 layersT2 : 2 layers

Discussion on OLSNC

• Optimal result for LSNC • High Computing Complexity E.g. 15 receivers, 5 layers, worst

cast execution time is over 1 hour

• A time efficient algorithm is needed

Sub-optimal Layer Separated Network Coding

Main Idea: 1) Allocate the bandwidth for each layer from low to high, with the objective of maximizing the aggregated maxflows

of receivers for rest higher layers. 2) Achieve the multicast rate for each layer with the bandwi

dth allocated by existed network coding algorithm.

*

1

( , ) ( , )

{ |( , ) } { |( , ) }

0

, , , 0 (1)

t

i j j i

lt

v T

l t l tij ji

j v v E j v v E

i t

j

Max f

Subject to

x x

for v T v T t i i

x

0

( , )

{ |( , ) }

( , )0

{ |( , ) }

( , )

( 1, ) ( 1, )

{ |( { |( , ) }

(2)

(3)

, ( , )

0

j t

j

ij

i j i

l tt l t

j v v E

l tj l t

j v v E

l l tij t i j

l t l tij ji

j v j v v E

r for v T

x r for v T

Y x for v T v v E

x x

0

, ) }

* *

( 1, ) 1 *

{ |( , ) }

( 1, ) 1 *0

{ |( , ) }

1 ( 1, ) *

, , , 0 (4)

(5)

(6)

, ( ,

j

j t

j

ij

v E

i t

l t ljt t t

j v v E

l t lj t t

j v v E

l l tij t i

for v T v T t i i

x f for v T

x f for v T

Z x for v T v v

1

)

( , ) ( , ) (7)ij ij

j

l li j i j

E

Y Z C v v for v v E

Performance Evaluation

Simulation Environments V= {v0, v1,…v10}, R={v1, v2,…v10} Two topologies: E1={(vi,vj)| i < j }, E2=((vi,vj)| 0 < j−i ≤ 2 } Two layer rate allocation schemes: Flat and Exponential

Scheme Performance metrics:

, ( )

, ( )

( )

( )i i

i i

iv R OL v k

ki

v R OL v k

AC v

LRROL v

( )

( )i

i

iv R

iv R

AC v

LRROL v

AC(Vi) is the actual number of layers received by viOL(vi) is the maximum number of layers permitted by maxflow

0

0. 2

0. 4

0. 6

0. 8

1

LRR1 LRR3 LRR5

ROME OLSNC SLSNC

0

0. 2

0. 4

0. 6

0. 8

1

LRR1 LRR3 LRR5

ROME OLSNC SLSNC

Simulation Results

E1, Flat Scheme E1, Exponential Scheme

Simulation Results

0

0. 2

0. 4

0. 6

0. 8

1

LRR1 LRR3 LRR5

ROME OLSNC SLSNC

0

0. 2

0. 4

0. 6

0. 8

1

LRR1 LRR3 LRR5

ROME OLSNC SLSNC

E2, Flat Scheme E2, Exponential Scheme

The advantage is more obvious in E1, with Exponential Scheme.

Simulation Results cont

E=E1

Flat Rate

E=E1,

Exponential

E=E2,

Flat Rate

E=E2,

Exponential

ROME 0.9361 0.9560 0.7276 0.9573

OLSNC 1.0000 1.0000 0.9987 0.9982

SLSNC 1.0000 1.0000 0.9941 0.9923

The comparison of LRR

Future Works

• In undirected networks

• Distributed Network Coding Scheme • Fairness problem

• Layered P2P Streaming Using Network Coding

Thank You