optimizing index allocation for sequential data broadcasting in wireless mobile computing

40
Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing Ming-Syan Chen, Senior Member, IEEE, Kun-Lung Wu, Member, IEEE Computer Society, and Philip S. Yu, Fellow, IEEE M9129022 郭郭郭

Upload: ulric-delaney

Post on 01-Jan-2016

15 views

Category:

Documents


0 download

DESCRIPTION

Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing. Ming-Syan Chen, Senior Member, IEEE, Kun-Lung Wu, Member, IEEE Computer Society, and Philip S. Yu, Fellow, IEEE. M9129022 郭文漢. Outline. Introduction Preliminaries - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Optimizing Index Allocation for Sequential Data Broadcasting in Wireless

Mobile Computing

Ming-Syan Chen, Senior Member, IEEE, Kun-Lung Wu, Member, IEEE Computer Society, and Philip S. Yu, Fellow, IEEE

M9129022 郭文漢

Page 2: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Outline

1. Introduction

2. Preliminaries

3. Index Allocation for Skewed Data Access

4. Optimal Order for Sequential Data Broadcasting

Page 3: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Introduction

背景

建立 index tree

Algorithm CF Algorithm VF

Optimal orderfor sequential

data broadcasting

解決方法 效益

節省電力

Algorithm ORD

舊方法問題問題

不使用Data Access Skew

有限電力

Page 4: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Introduction

A mobile client to be able to operate in two different modes: doze mode and active mode.

The structure of an index tree determines the index probing scenario to switch between the doze and the active modes for data access under such an indexed broadcasting.

Data Access Skew : The access frequencies of different data records are usually different from one another.

Page 5: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Introduction

I

a1 a2 a3

R1 R2 R3 R4 R5 R6 R7 R8 R9

I a1 R1 R2 R3 a2 R4 R5 R6 a3 R7 R8 R9

Indexed broadcastingIndex tree

Index probing scenario to data record R5

Page 6: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Preliminaries

A mobile client is assumed to use selective tuning to listen to indexed sequential data broadcasting.

Tuning time : The amount of time spent by a client to listen to the channel.

Access time : The time elapsed from the time a client wants an identified record to the time that record is downloaded by the client.

Page 7: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Preliminaries

Probe wait : The time from the point a client tunes in to the point when the first index is reached.

Bcast wait : Time duration from the point the first index is reached to the point the required record is obtained.

Page 8: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Preliminaries

Tuning timeClient

I

a1 a2 a3

R1 R2 R3 R4 R5 R6 R7 R8 R9

Probe wait

Bcast wait

Page 9: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Index Allocation For Skewed Data Access

1. Imbalanced Index Tree Construction for Fixed Fanouts

2. Employing Variant Index Fanouts to Minimize Index Probes

3. Experimental Results on Index Allocation

Page 10: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Imbalanced Index Tree Construction for Fixed Fanouts

Algorithm CF will reduce the number of index probes for hot data while allowing more probes for cold data.

Algorithm CF : Use access frequencies to build an index tree with a fixed fanout d.

Page 11: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Algorithm CF (bottom up manner)

Step 1 : Every single node labeled with the corresponding access frequency.

Step 2 : Attach the d subtrees with the smallest labels to a new node. Label the resulting subtree with the sum of all labels from its d child subtrees.

Step 3 : n=n-d+1. If n=1 stop else goto Step2

Page 12: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Algorithm CF

R10.4

R20.4

R30.05

R40.05

R50.02

R60.02

R70.02

R80.02

R90.02

a30.06

R10.4

R20.4

R30.05

R40.05

R50.02

R60.02

R70.02

R80.02

R90.02

a30.06

R10.4

R20.4

R30.05

R40.05

R50.02

R60.02

R70.02

R80.02

R90.02

a20.09

a30.06

R10.4

R20.4

R30.05

R40.05

R50.02

R60.02

R70.02

R80.02

R90.02

a20.09

a10.2

Page 13: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Algorithm CF

a30.06

R10.4

R20.4

R30.05

R40.05

R50.02

R60.02

R70.02

R80.02

R90.02

a20.09

a10.2

I

I R1 R2 a1 R3 a2 R4 R5 R6 a3 R7 R8 R9

Corresponding data broadcasting sequence

Page 14: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Cost Model

Theorem 1: Given a fixed index fanouts, the average

number of index probes is minimized by using the index tree constructed by algorithm CF.

Cost model

ni RPatha

jiIjd

ij

adRTC1 )(

)()Pr()(

Page 15: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Cost Model

ni RPatha

jiIjd

ij

afRTC1 )(

)()Pr()(

iaindex an probing ofcost the:)( iaf

iR record data root to thefrom nodesindex ofset the:)( iRPath

))(()( ii adgaf function increasinglly monotonica a:(.)g

)())(()( iii adadgaf

index. that offanout the toalproportion is

index an probing ofcost theng,broadcasti sequentialUnder

Page 16: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Cost Model

9333)()()()( 21)( 5

adadIdadRPatha

j

j

Page 17: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Cost Model

28.4)(12.4)(

05.402.0*4505.0*154.0*6)(

24

3

ITCTC

TC

dId

Id

Page 18: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Employing Variant Index Fanouts to Minimize Index Probes

An efficient heuristic algorithm VF to build an index tree with variant fanouts.

We want data records to stay as close to the root as possible.

Algorithm VF strikes a compromise between these conflicting factors( larger fanouts) and minimizes the average cost of index probes.

Page 19: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Employing Variant Index Fanouts to Minimize Index Probes

1 2

Lemma 1.

Suppose that node r has m child nodes, , , ..., ,

which are sorted according to descending order of Pr( ),

1 , i.e. Pr( ) Pr( ) if and only if j .Then, the

average cost of index

m

j

j k

h h and h

h

j m h h k

1 2

1 1

probes can be reduced by grouping

nodes , , ..., and and attaching them under a new

child node if and only if

( 1) Pr( ) Pr( )

i i m

j jj i i j m

h h h

m i h h

Page 20: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Employing Variant Index Fanouts to Minimize Index Probes

( )

1 ( )

1 1 ( )

Pr( )( ( ) )

Pr( )( ( ) ) Pr( )( ( ) )

p j

p PT h p jj

BE j pj m a PT h

j p j pj i a i j m a PT h

C h d a m

h d a m h d a m

Page 21: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Employing Variant Index Fanouts to Minimize Index Probes

1 ( )

1 ( )

Pr( )( ( ) ( 1))

Pr( )( ( ) ( 1 ))

p j

p j

AF j pj i a PT h

j pi j m a PT h

C h d a i

h d a i m i

Page 22: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Employing Variant Index Fanouts to Minimize Index Probes

1 1

1 1

( 1 ) Pr( ) Pr( )

( 1 ) Pr( ) Pr( )

AF BE j jj i i j m

j jj i i j m

C C i m h h

i m h h

Page 23: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Algorithm VF (top down manner)

1 2

1 2

Algorithm VF:

Step 1:Assume that , , ..., and have been sorted

according to descending order of Pr( ), 1 .

Step 2:Partition( , , ..., ).

Step 3:Report the resulting index tree.

n

j

n

R R R

R j n

R R R

Page 24: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Algorithm VF

* *

1 2

1 1

* *

1, 2

*

1 2

Procedure Partition( , , ..., ):

1.Let ( ) ( 1) Pr( ) Pr( ).

Determine such that ( ) max ( )

2.If y(i ) 0, then return.

3.Attach nodes , , ..., under a

m

j jj i i j m

i m

mi i

h h h

y i m i h h

i y i y i

h h h

* *

*

*

1 2

1 2

1 2 1

new

index node in the index tree.

4.Partition( , , ..., ).

5.Insert into the ordered list ( , , ..., )

and relabel them as ( , , ..., ) according

to descending ord

x

mi i

x i

i

h

h h h

h h h h

h h h

*

*

1 2 1

er of Pr( ), 1 1.

6. artition( , , ..., ).

7.Re .

j

i

h j i

P h h h

turn

Page 25: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Algorithm VF

Page 26: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Algorithm VF

Page 27: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Algorithm VF

Page 28: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Algorithm VF

Page 29: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Algorithm VF

R1 R2 R3 R4 R5 R6 R7 R8 R9 R10 R11

4 4 5 5 7 10 10 14 14 14 14

08.5)(

625.5)(

2.5)()Pr()(

3

)(1112

IV

Id

RPathaj

ii

Id

TC

TC

adRTCjj

Page 30: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Experimental Results on Index Allocation

s.frequencie access

oforder descendingin sorted items

data are ,1, andfactor skew

theis where,)()Pr(Let

on.distributi Zipfby the modelled are

items data of frequences access The

niRn

inR

i

i

h

h

Page 31: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Experimental Results on Index Allocation

Page 32: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Optimal Order for Sequence Data Broadcasting

1. Ordering Broadcasting Data to Minimize Data Access Time

2. Experimental Results on Order of Broadcasting

3. Remarks

Page 33: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Ordering Broadcasting Data to Minimize Data Access Time

Page 34: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Ordering Broadcasting Data to Minimize Data Access Time

Page 35: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Algorithm ORD

Page 36: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Algorithm ORD

Page 37: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Algorithm ORD

Page 38: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Experimental Results on Order of Broadcasting

Page 39: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

Remarks

Algorithm Complexity Operation

CF sorting

VF recursive

ORD sorting

)log( nnO

)log( nnO

)log( 2 nnO

Page 40: Optimizing Index Allocation for Sequential Data Broadcasting in Wireless Mobile Computing

謝謝