1 Adaptive resource management with dynamic reallocation for layered multimedia on wireless mobile communication net work Date ： 2005/06/07 Student ： Jia-Hao.

1

Adaptive resource management with dynamic reallocation for layered multimedia on wireless mobile communication net work

Date： 2005/06/07

Student： Jia-Hao Xu

Advisor： Kai-Wei Ke

2

Outline

Motivation Background Multimedia Streaming Service Call admission Control for Single-Media Call admission Control for Multimedia Numerical Results Conclusion

3

Motivation

We want to find an optimal call admission control policy for wireless multimedia streaming service.

Main issues： To promote Wireless bandwidth resources utilizati

on. To guarantee handover call blocking probability.

4

Background

Wireless communication network introduction Wireless resource allocation method MDP (Markov Decision Process) introduction

5

Wireless communication network introduction

6

Wireless Resource allocation method – 2G： GSM (Global System for Mobile)

It uses Frequency Division Multiple Access (FDMA) and Time Division Multiple Access (TDMA) to access air interface.

Divide carrier into multi TDMA frame (4.615ms) and each frame consist of eight time slots (577us).

It uses 51-Multiframe (51 TDMA frame).

7

Wireless Resource allocation method – 2G： GSM (Global System for Mobile)

The unit of resources allocation is time slot.

8

Wireless Resource allocation method – 2.5G： GPRS (General Packet Radio Service)

The difference between GSM and GPRS is GPRS using 52-Multiframe (52 TDMA frame).

There are 4 IDLE frames and 12 Radio Blocks (each block contains 4 continuous frame).

9

Wireless Resource allocation method – 2.5G： GPRS (General Packet Radio Service)

The unit of resources allocation is radio block.

10

Wireless Resource allocation method – 3G：WCDMA (Wideband Code Division Multiple Access) It uses Code Division Multiple Access (CDMA)

to access air interface and Direct Sequence Spreading Code (DSSC) modulation technique.

Each user allocate a orthogonal spreading code produced by OVSF Code (Orthogonal Variable Spreading Factor Code) technique.

The transmission depends on Spreading Factor.

11

Wireless Resource allocation method – 3G：WCDMA (OVSF Tree)

The unit of resources allocation is SF (4~256).

12

MDP introduction

A Markov Decision Process is just like a Markov Process, except the transition depends on the action at each time step.

A system’s state has various options (actions) to choose from, and each option associates a “reward”.

The goal is to find a optimal policy, which specifies which action to take in each state, so as to maximize rewards. (solved by LP)

13

Outline

Motivation Background Multimedia Streaming Service Call admission Control for Single-Media Numerical Results Conclusion

14

Streaming

Streaming is a process of playing a file while it is still downloading. And, it uses buffering techniques.

When a large media file (audio, video, etc.) is broken into smaller pieces so it can viewed or heard immediately. This avoids the wait for the whole file to be downloaded first.

It doesn’t require low delay but low jitter and media synchronization.

15

Multimedia application

There are many application using this technique, such as video phone, video-on-demand and interactive video games.

These services consume lots of bandwidth resources.

Layered encoding (multirate and adaptive)：A technique to convert a file into a compressed, streaming format. For example, H.263, MPEG-2, and MPEG-4

16

Streaming service architecture

RTSP (TCP)RTP (UDP)HTTP (TCP)FTP(TCP)

Networkcomponents

Proxy

RTSP/RTPSDP

HTTPFTP

Content creation machines

IP

Streaming servers

Webservers

IP

SMIL,Images,

textStreams

IP network

17

Our streaming architecture

Base station (do

some things)

System Web

Term

inal user

（traveler

）

Streaming server

Wireless transmission

Select multimedia file

Get file from streaming server

Return multimedia file

For example, adaptive and reallocation resources

18

Main issuse

Design a dynamic and adaptive resource allocation strategy to

1. Lower Blocking probabilitynew and handover call (especially)

2. Increase resources utilization

%

allocated bandwidth resourcesU

total bandwidth resources

19

Outline


20

Network model

Base station

21

Network model (cont.)

Each base station has a total of “C” bandwidth resources and support “k” different types of encoding method (means “k” different rates).

We assume that new arriving calls in a cell follow a Poisson process.

Both call session time and call dwell time are exponentially distributed and are independent from cells to cells

22

Network model (cont.)1 Assume the arrival rate of new call is , the arrival

rate of handover call is , the mean session time of

a call is 1/ and the mean dwell time is 1/ .

2 Any user uses bandwidth ,

n

h

s h

ib b

、

、 1 1

1

,..., ,..., | ,

1 1

3 System state present as ( ,..., ,..., ) ,and means the

number of users who use bandwidth ,1

4 We get state space of the system :

j k j j

i

i k i

i

b b b b b

for j k

x x x x x

b i k

、

、

1

{ | and 0 for all 1 }k

i i ii

x b x C x i k

23

MDP-based Call Admission Control

1 2 31 2 1, , 2, , 3, ,

1

1 At the decision epoch, the action space containing all

possible decisions and reaction methods

( , , , , ) : 0,1 , 1,2

where

0 or 1 : reject or accept

x p x p x p i

A

A d d a a a d i

d

、

1 2 3

2

1, , 2, , 3, ,

new call.

0 or 1 : reject or accept handover call.

, and are the reaction for a new call arrival,

a handover call arrival and a call departure.

x p x p x p

d

a a a

24

MDP-based Call Admission Control (cont.)

3, , 3, ,

We can get the action space of state ,

: 0 1, 2

3 Whether the action is for new or handover call, the state

will switch to a new state

i

x

x i i x p x p

i

x A

A a A d if a i and a

x

x a

2、

、

3

, ,

3, , ,

, 1, 2

4 When a call departures from state ,still switch to a new state

ix p

x g p

i

x

x a

、

25


1

1 21 1

The average sojourn time in state when action is chosen :

( , ) ( ) ( )

The transition probability that at next decision epoch the

system will be i

k k

n h i s i hi i

x a

x a d d x x

5、

6、

1 21 1

11 1

n state given that the action is chosen

by the current state is :

( ) ( , ) , 1

( ) ( | , ) , 1

0 ,

k k

n h i ii i

kk k

i s hi i i

i i

y a

x

d d x a if y x and y x

xP y x a if y x and y x

otherw

ise

26


1

The expected reward obtained until the next decision epoch

given that the action is chosen by the current state is :

( , )

8 We define decision variable , as the long-run

k

i ii

a x

r x a b x

u x a

7、

、

the set of ,

collectively determines the policy

Searching for t

fracton

time at which the s

he optimal policy is equivalent

tate

to finding

those decision va

making action , and

.

9

x a u x a

、. This can be done by

formulation with the objective o

riables for all states

linear prog n

maximizing long-run netwo

rammi

rk re

ng

wa

(LP)

rd.

27


2

Maximize :

, , ,

Subject to :

, , 1

, ( | , ) , y

1 , ,

, 0 , ,

x

x

y x

x

x a A

x a A

a A x a A

BHx a A

x

U r x a x a u x a

x a u x a

u y a P y x a u x a for

d x a u x a P

u x a x a A

28

Three Strategies for

Adaptive Only Adaptive and Dynamic Reallocation Finally Adaptive and Dynamic Reallocation Anytime

1 2 31, , 2, , 3, ,, ,x p x p x pa a a

29

Adaptive Only (AO) --

Remaining resource ?

New state

Call blocked

Originating call

NO

YES Adaptive resource allocation

1b

11, ,x pa22, ,x pa

X = (0,0) and call arrival => (1,0) or (0,1)

30


1

1

1 1 1

1, , 1 1

1

1, , 1 1 1

1

,..., , 0

,..., ,..., (0,...,1,...,0) |

... ( 1) ... , 1

1

x p k

p k

x p p p k k

a x x x if d

x x x

a x b x b x b C if d

and p k

11, ,x pa22, ,x pa

2

1

2 2 2

2, , 1 2

1

2, , 1 1 2

2

,..., , 0

,..., ,..., (0,...,1,...,0) |

... ( 1) ... , 1

1

x p k

p k

x p p p k k

a x x x if d

x x x


and p k

31


New state

Call departure

Resources release

33, ,x pa

32


3

3

3

3

3, , 11

1

3, ,13

,..., , 0

,..., 1,..., |, 0

0 1

k

x p k ii

kp k

x p iip

a x x x if x

x x xa if x

x and p k

33, ,x pa

33

Adaptive Only (AO) -- example

1 23 , 2 , , 1, 2C k b b

34

Adaptive and Dynamic Reallocation Finally (ADRF) --

Remaining resource ?

New state

Resource reallocation

Call blocked

Originating call

NO

YES

YES

NO

Adaptive resource allocation

1b

11

/ ?k

ii

x C b

11, ,x pa22, ,x pa

X = (1,2) => (0,3)or(1,2)or (2,1)or(3,0)

35


11, ,x pa

1

1

1 1 1

1

1

1, , 1 1

1

1, , 1 1 1 1

1

' ' '1

1, ,

,..., , 0

,..., ,..., (0,...,1,...,0) |

... ( 1) ... , 1 resources

1

,..., ,...,

x p k

p k

x p p p k k

p k

x p

a x x x if d

x x x

a x b x b x b C if d and b

and p k

x x x

a

1 1

' ' '1 1 1 1

1

(0,...,1,...,0) |

... ( 1) ... , 1 resources

1

p p k kx b x b x b C if d and b

and p k

<

36


2

1

2 2 2

1

2

2, , 1 2

1

2, , 1 1 2 1

2

' ' '1

2, ,

,..., , 0

,..., ,..., (0,...,1,...,0) |

... ( 1) ... , 1 resources

1

,..., ,...,

x p k

p k

x p p p k k

p k

x p

a x x x if d

x x x

a x b x b x b C if d and b

and p k

x x x

a2 2

' ' '1 1 2 1

2

(0,...,1,...,0) |

... ( 1) ... , 1 resources

1

p p k kx b x b x b C if d and b

and p k

22, ,x pa

37


Call departure

Resources release

Resources reallocationNew state

33, ,x pa

38

Adaptive and Dynamic Reallocation Finally (ADRF) -- 33, ,x pa

3

3 1

3

3

3, , 11

' '1

' '3, , 1 1

1

3

,..., , 0

,..., 1,..., ( ,..., ) |

... , 0

0 1

k

k

x p k ii

p kk

x p k k ii

p

a x x x if x

x x x x x

a x b x b C if x

x and p k

39

Adaptive and Dynamic Reallocation Finally (ADRF) -- example 1 23 , 2 , , 1, 2C k b b

40

Adaptive and Dynamic Reallocation Anytime (ARDA) --

New state

Resource reallocation

Call blocked

Originating call

YES

NO

Adaptive resource allocation

11

/ ?k

ii

x C b

11, ,x pa22, ,x pa

41


1

1

1 1 1

1, , 1 1

' ' '1

' ' '1, , 1 1 1

1

,..., , 0

,..., ,..., (0,...,1,...,0) |

... ( 1) ... , 1

1

x p k

p k

x p p p k k

a x x x if d

x x x


and p k

2

1

2 2 2

2, , 1 2

' ' '1

' ' '2, , 1 1 2

2

,..., , 0

,..., ,..., (0,...,1,...,0) |

... ( 1) ... , 1

1

x p k

p k

x p p p k k

a x x x if d

x x x


and p k

11, ,x pa22, ,x pa

42


Call departure

Resources release

Resources reallocationNew state

33, ,x pa

43


3

3 1

3

3

3, , 11

' '1

' '3, , 1 1

1

3

,..., , 0

,..., 1,..., ( ,..., ) |

... , 0

0 1

k

k

x p k ii

p kk

x p k k ii

p

a x x x if x

x x x x x

a x b x b C if x

x and p k

33, ,x pa

44

Adaptive and Dynamic Reallocation Anytime (ARDA) -- example 1 25 , 2 , , 1, 2C k b b

45

Outline


46

Numerical results – Adjust Offered Load

Normalized

offered load

n hk

s h

b

C

47


0

0.2

0.4

0.6

0.8

1

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8

Normalized Offered Load

Util

izat

ion

AO ADRF ADRA

48


0

0.002

0.004

0.006

0.008

0.01

0.012

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8

Normalized Offered Load

Blo

ckin

g Pr

obab

ility

AO ADRF ADRA

49

Numerical results – Adjust handover blocking probability upper bound

50


0

0.2

0.4

0.6

0.8

1

0 0.005 0.01 0.015 0.02 0.025

Handover Call Blocking Probability Upper bound

Util

izat

ion

AO ADRF ADRA

51


0

0.005

0.01

0.015

0.02

0.025

0 0.005 0.01 0.015 0.02 0.025

Handover Call Blocking Probability

Util

izat

ion

AO ADRF ADRA

52

Numerical results – Adjust new and handover call ratio

53


0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0 0.2 0.4 0.6 0.8 1 1.2

Handover Call Ratio

Util

izat

ion

AO ADRF ADRA

54


0

0.002

0.004

0.006

0.008

0.01

0.012

0 0.2 0.4 0.6 0.8 1 1.2

Handover Call Ratio

Blo

ckin

g Pr

obab

ility

AO ADRF ADRA

55

Why ADRF equals to ADRA ?

Look static state probability：

Probability ( , ) ( , )

3 , 1, 2 , , 0.002,0.005

Normalized offered load = 0.8

xx a A

s h

x a u x a

C B

56

Why ADRF equals to ADRA ? (cont.) Look taking which action：

5 , 1, 2C B

57

Conclusion

We propose three call admission control policy : AO, ADRF and ADRA

Through numerical analysis, we find that AO method is the worst strategy, ADRF and ADRA are the best strategy in single-media streaming service.

1 Adaptive resource management with dynamic reallocation for layered multimedia on wireless mobile communication net work Date ： 2005/06/07 Student ： Jia-Hao.

Documents

lp slide

unit of resources allocation

streaming streaming

multimedia application

orthogonal spreading

factor code technique

layered multimedia

markov process