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
Dec 19, 2015
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
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
Motivation Background Multimedia Streaming Service Call admission Control for Single-Media Numerical Results Conclusion
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
MDP-based Call Admission Control (cont.)
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
MDP-based Call Admission Control (cont.)
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
MDP-based Call Admission Control (cont.)
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
Adaptive Only (AO) --
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
a x b x b x b C if d
and p k
32
Adaptive Only (AO) --
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
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
Adaptive and Dynamic Reallocation Finally (ADRF) --
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
Adaptive and Dynamic Reallocation Finally (ADRF) --
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
Adaptive and Dynamic Reallocation Finally (ADRF) --
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
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
Adaptive and Dynamic Reallocation Anytime (ARDA) --
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
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
a x b x b x b C if d
and p k
11, ,x pa22, ,x pa
42
Adaptive and Dynamic Reallocation Anytime (ARDA) --
Call departure
Resources release
Resources reallocationNew state
33, ,x pa
43
Adaptive and Dynamic Reallocation Anytime (ARDA) --
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
45
Outline
Motivation Background Multimedia Streaming Service Call admission Control for Single-Media Numerical Results Conclusion
47
Numerical results – Adjust Offered Load
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
Numerical results – Adjust Offered Load
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
50
Numerical results – Adjust handover blocking probability upper bound
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
Numerical results – Adjust handover blocking probability upper bound
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
53
Numerical results – Adjust new and handover call ratio
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
Numerical results – Adjust new and handover call ratio
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