rani nalamaru department of computer science ball state university rani nalamaru department of...

31
RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY Efficient Transmission of Stored Video for Improved Management of Network Bandwidth

Upload: oscar-dennis

Post on 31-Dec-2015

215 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY Efficient Transmission

RANI NALAMARU

DEPARTMENT OF COMPUTER SCIENCE

BALL STATE UNIVERSITY

RANI NALAMARU

DEPARTMENT OF COMPUTER SCIENCE

BALL STATE UNIVERSITY

Efficient Transmission of Stored Video for Improved

Management of Network Bandwidth

Page 2: RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY Efficient Transmission

Overview of presentation

Introduction Background Problem Statement

The New VP Algorithm Evaluation of OBA, Optimal and VP Algorithms Summary and Future work

Page 3: RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY Efficient Transmission

IntroductionNetwork video

Many emerging applications

» Entertainment, Distance learning, Catalogue browsing

etc.

Video packet

Client

Video ServerNetwork

Client

Storage

Page 4: RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY Efficient Transmission

Introduction

Networking challenges for video

Huge bandwidth requirement if no compression With compression traffic is bursty

» Bursty traffic complicates network management

Goal: Efficient transmission of high quality stored streaming

video

Page 5: RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY Efficient Transmission

Video compression and burstiness

Burstiness can occur due to:» Type of frames used in encoding» Background changes or changes in scene content

Introduction

0

5000

10000

15000

20000

25000

30000

35000

1

539

1077

1615

2153

2691

3229

3767

4305

4843

5381

5919

6457

6995

7533

8071

8609

Frame Number

Fram

es S

ize

(in

byte

s)

Frame sizes of a stored video

Page 6: RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY Efficient Transmission

Background

Transmission plan Pre-calculated schedule to transmit a video file

Mechanism to smooth the bandwidth requirement

0 3300 224243301 10400 14929

Start End Bandwidth

Read video frames from disk

Implement the transmission plan

Transmission of frames to network

Decode and displayof frames

Buffering of framesin client buffer

Receive framesfrom network

Server Client

Network

Page 7: RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY Efficient Transmission

Background

Given the parameters : Frame sizes for n frames Client buffer size b

Constraints at the client buffer Avoid buffer underflow Avoid buffer overflow

Have all video frames in advance» Knowledge of frame sizes

Goal: Find a transmission plans with minimum number of rate changes and minimized sum of rate variation

iS

n

jii Sf

1

n

jii SbF

1

Work-ahead smoothing

cum

ulat

ive

fram

e si

ze

timeS 1

S 2

S 3

S 4

S 5

S 6

S 7

bandwidthchanges

b

f i

F i

Page 8: RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY Efficient Transmission

BackgroundOptimal Bandwidth Allocation (OBA) algorithm (1995)

Developed by Feng, Jahanian, and Sechrest (Univ. of Michigan)

Goal of OBA algorithm is to develop a transmission plan with

» smallest peak bandwidth

» largest minimum bandwidth

» fewest possible changes in bandwidth (rate changes)

Page 9: RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY Efficient Transmission

BackgroundOptimal algorithm (1996)

Developed by Salehi, Kurose, and Towsley (Univ. of Mass.)

Goal of Optimal algorithm was to develop a transmission plan with

» smallest peak bandwidth

» least variation between bandwidth changes (rate variation)

Page 10: RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY Efficient Transmission

Problem Statement

Problems with existing algorithms

Buffer sizes in the range of 20-30Mbytes are required

Retains the VBR property of stored video

Time complexity is of the order of O(N logN) and O(N 2)

» N is the number of frames

Page 11: RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY Efficient Transmission

Problem StatementPossibility of improvement

When to change transmission rate ?

Cum

ulat

ive

Fram

esiz

e

Frame Number

Optimal

OBA

0

We wish to usebest of both

Page 12: RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY Efficient Transmission

Visibility Polygon (VP) Algorithm

Solution - VP algorithm

Develop an algorithm based on visibility concept

» Developed by Subhash Suri ( John Hopkins, 1986)

What is visibility ?

Set of points that are visible from a given point in a region

visible to a

a b

c

not visible to a

Page 13: RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY Efficient Transmission

Visibility Polygon (VP) Algorithm

Steps in VP algorithm

1) Given frame sizes and client buffer size b. We construct the feasible region P.

iS

n

jii Sf

1

n

jii SbF

1

cum

ulat

ive

fram

e si

ze

time

FeasibleRegion( P )

f i

F i

b

Page 14: RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY Efficient Transmission

Visibility Polygon (VP) Algorithm

Steps in VP algorithm

2) Triangulate the feasible region P, let T represent the triangulation of P.

cum

ulat

ive

fram

e si

ze

time

FeasibleRegion( P )

f i

F i

Tb

Page 15: RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY Efficient Transmission

Visibility Polygon (VP) Algorithm

Steps in VP algorithm

3) Construct the dual graph G of triangulated polygon.cu

mul

ativ

e fr

ame

size

time

FeasibleRegion( P )

T

G

Page 16: RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY Efficient Transmission

Visibility Polygon (VP) Algorithm

Steps in VP algorithm

4) Identify the shortest path from first frame to last frame.

5) Compute the windows, from which transmission plan is obtained.

cum

ulat

ive

fram

e si

ze

time

Windows

Page 17: RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY Efficient Transmission

Visibility Polygon (VP) Algorithm

Complexity of VP algorithm

Triangulation ------------------------------------ O(N) Dual Graph Construction ---------------------------- O(N) Breadth First Search ------------------------------- O(N)

Visibility Polygon & Windows computation ------

Hence VP algorithm takes linear time

NONOi

i

An improvement over the previous algorithms which are O(N logN) and O(N2)

Page 18: RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY Efficient Transmission

Evaluation

Comparison of OBA, Optimal and VP Algorithms

Simulation model

Use trace files of representative videos

Parameters for evaluation Peak-rate bandwidth

Number of rate changes

Variation between rate changes

Time complexity

Page 19: RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY Efficient Transmission

Evaluation

Peak-rate bandwidth

Peak-rate bandwidth

0

5000

10000

15000

20000

25000

30000

35000

1

539

1077

1615

2153

2691

3229

3767

4305

4843

5381

5919

6457

6995

7533

8071

8609

Frame Number

Fram

es S

ize

(in

byte

s)

Page 20: RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY Efficient Transmission

Evaluation

Rate changes and variation between rate change

0

5000

10000

15000

20000

25000

30000

35000

1

539

1077

1615

2153

2691

3229

3767

4305

4843

5381

5919

6457

6995

7533

8071

8609

Frame Number

Fram

es S

ize

(in

byte

s)

Intervals

Variation

Page 21: RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY Efficient Transmission

Evaluation

Time complexity

Measure the number of seconds for calculating transmission plan

Page 22: RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY Efficient Transmission

Evaluation

Experimental setup

Java Simulation Program

Video frames retrieved from server storage

Transmission plan

Page 23: RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY Efficient Transmission

Evaluation

Validation of simulation model Feng's OBA algorithm Our OBA algorithm

Maximum Bandwidth 24069 bytes/sec 24069 bytes/sec

Number of Changes 20 17

Variation Between Changes 363 % 306 %

0

5000

10000

15000

20000

25000

30000

0 50000 100000 150000

Frame Number

Tra

nsm

issi

on P

lan(

Byt

es/s

ec) Feng's OBA

Our OBAConservative

results

Page 24: RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY Efficient Transmission

Evaluation

Inputs Videos were selected to be representative with respect to length and subject material

Video Trace Length Bit Rate Max Frame Min Frame Std Dev

Beauty & Beast 80 min 3.0 Mbps 30367 bits 2701 bits 3580 bits

Big 102 3.0 23485 1503 2366

Crocodile Dundee 94 2.6 19439 1263 2336

Extra-Terrestrial 110 1.5 14269 1153 1840

Page 25: RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY Efficient Transmission

Evaluation

Peak-rate bandwidth

10000

15000

20000

25000

Beauty&Beast Big Dundee ET

OBAOptimalVP

8 %

3.7 %

OBA

Optimal

Video Trace OBA algorithm Optimal smoothing VP algorithm

Beauty & Beast 24069 bytes/sec 23124 bytes/sec 22424 bytes/sec

Big 15474 15516 15000

Crocodile Dundee 13321 14401 13176

Extra-Terrestrial 21731 18564 18442

Page 26: RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY Efficient Transmission

Evaluation

Number of rate changes

19 %

8.3 %

OBA

Optimal

0

5

10

15

20

25

30

Beauty&Beast Big Dunde ET

Video Files

Nu

mb

er

of

Ch

an

ge

s re

qu

ire

d

New AlgorithmOBAOptimal

Video Trace OBA algorithm Optimal smoothing VP algorithm

Beauty & Beast 17 15 15

Big 21 19 18

Crocodile Dundee 16 14 12

Extra-Terrestrial 28 24 21

Page 27: RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY Efficient Transmission

Evaluation

Amount of variation

15.3 %

9.6 %

OBA

Optimal

5

55

105

155

205

255

305

Beauty & Beast Big Dundee ET

Video File

% V

ari

ati

on

be

twe

en

ch

an

ge

s

New Algorithm

OBA

Optimal

Video Trace OBA algorithm Optimal smoothing VP algorithm

Beauty & Beast 306 % 280 % 262 %

Big 16 13 13

Crocodile Dundee 18 21 14

Extra-Terrestrial 17 15 10

Page 28: RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY Efficient Transmission

Evaluation

Time complexity

73.6 % OBA 3.8 % Optimal

Video Trace OBA algorithm Optimal smoothing VP algorithm

Beauty & Beast 4.1 sec 1.1 sec 1.1 sec

Big 5.4 1.3 1.4

Crocodile Dundee 4.7 1.2 1.3

Extra-Terrestrial 5.5 1.4 1.4

Page 29: RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY Efficient Transmission

Evaluation

What does all this mean to end users ?

If VP algorithms is used

If other algorithms are used

Video Server

Clients

Page 30: RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY Efficient Transmission

Summary and future work

Summary Problems with efficiently transmitting stored (compression) video

Reviewed OBA and Optimal algorithms

New VP algorithm proposed

Simulation results showed VP algorithm has better performance to its predecessors

Page 31: RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY RANI NALAMARU DEPARTMENT OF COMPUTER SCIENCE BALL STATE UNIVERSITY Efficient Transmission

Summary and future work

Future work

To implement VP algorithm on an actual video server

To study issues of multicast support of VP algorithm