An Empirical Study of RealVideo Performance Across the Internet

37
An Empirical Study of RealVideo Performance Across the Internet Yubing Wang, Mark Claypool and Zheng Zuo http://perform.wpi.edu/real-tracer

description

An Empirical Study of RealVideo Performance Across the Internet. Yubing Wang, Mark Claypool and Zheng Zuo http://perform.wpi.edu/real-tracer. Introduction. Streaming Audio and Video is growing Traditional Internet studies, but ….video is different Bandwidth and response time not enough - PowerPoint PPT Presentation

Transcript of An Empirical Study of RealVideo Performance Across the Internet

Page 1: An Empirical Study of RealVideo Performance Across the Internet

An Empirical Study of RealVideo Performance Across the Internet

Yubing Wang, Mark Claypool and Zheng Zuo

http://perform.wpi.edu/real-tracer

Page 2: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

Introduction

• Streaming Audio and Video is growing

• Traditional Internet studies, but ….video is different– Bandwidth and response time not enough

• Care about smooth rate

– Packet loss rate alone is not enough• Often use repair techniques

• RealVideo is big [Jup01]

– RealPlayer, MediaPlayer, QuickTime

Measure RealVideo on Internet

Page 3: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

Outline

• Introduction

• RealVideo

• Methodology

• Results

• Analysis

• Conclusions

Page 4: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

RealVideo Network Characteristics

• Buffering

• SureStream

• Scalable Video Technology

• Repair

Server

RTSP

Data: TCP or UDP

Page 5: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

Outline

• Introduction

• RealVideo

• Methodology

• Results

• Analysis

• Conclusions

Page 6: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

Methodology

• Build customized RealVideo player (RealTracer)– Uses RealPlayer core

– Software Development Kit (SDK)

– Records performance stats: frame rate, jitter ….

• Select diverse set of RealVideo servers and videos

• Solicit volunteers to run RealTracer– Many users to Many servers

– Friends, colleagues outside of Mass., U.S.

– rec.video newsgroup and end2end mailing list

– 2 weeks in June 2001

• Analyze results

Page 7: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

Outline

• Introduction

• RealVideo

• Methodology

• Results

• Analysis

• Conclusions

Page 8: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

TCP UDP

Results

• 11 servers in 8 countries

• 63 users in 12 countries

• 2855 video clips played

• 388 video clips watched and rated

• 10% chance clip unavailable

Page 9: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

Geographic Location of Servers and Clients

Page 10: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

Clips Played per User

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 20 40 60 80 100

Clips Played per User

Cu

mu

lati

ve

De

nsi

ty

.

Page 11: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

Clips Served per Country

126 184 240 260 294 297416

1075

0

200

400

600

800

1000

1200

Can

ada

Japa

n

Italy

Chi

na

Aus

tral

ia

Bra

zil

UK

US

Nu

mb

er

of

Clip

s

Page 12: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

Clips Played per Country

8 16 32 47 55 59 84 98 115 131 142

2100

0

500

1000

1500

2000

2500

Egyp

t

Indi

a

New

Zea

land

Rom

ania

UA

E

UK

Can

ada

Aus

tral

ia

Fran

ce

Ger

man

y

Chi

na US

Nu

mb

er

of

Clip

s

Page 13: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

0

200

400

600

800

1000

1200

VA WA ME TN CT NH CO IL TX CA WI DE MD MN NC FL MA

State Names

Nu

mb

er

of

Clip

s

Clips Played per State

Page 14: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

Outline

• Introduction

• RealVideo

• Methodology

• Results

• Analysis– Frame Rate – Jitter– Perceived Quality

• Conclusions

Page 15: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

Frame Rate

• Basic video performance unit– frames per second (fps)

• Guidelines [Rea00a]:

– 24-30 fps: full-motion video

– 15 fps: full-motion video approximation

– 7 fps: choppy

– 3 fps: very choppy

– Less than 3 fps: slide show

Page 16: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

Overall Frame Rate

00.1

0.20.30.40.5

0.60.70.8

0.91

0 5 10 15 20 25 30

Average Frame Rate (fps)

Cu

mu

lati

ve D

en

sit

y .

Page 17: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

Frame Rate vs. Network Configuration

Modem: 19% (550)DSL/Cable: 30% (874)T1/LAN: 51% (1468)

Page 18: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

Bandwidth vs. Network Configuration

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 100 200 300 400 500

Average Bandwidth (Kbps)

Cum

ulat

ive

Den

sity

56k Modem

DSL/Cable

T1/LAN

Page 19: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

Frame Rate vs. Transport Protocol

0

0.2

0.4

0.6

0.8

1

0 10 20 30

Frame Rate (fps)

Cu

mm

ula

tive

De

ns

ity

TCP

UDP

Page 20: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

Bandwidth vs. Transport Protocol

0

0.1

0.2

0.3

0.40.5

0.6

0.7

0.8

0.9

1

0 100 200 300 400 500 600

Average Bandwidth (Kbps)

Cu

mu

lati

ve

De

ns

ity

TCP

UDP

Page 21: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

Frame Rate vs. Server Geographic Location

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 5 10 15 20 25 30

Average Frame Rate (fps)

Cu

mm

ula

tive

De

ns

ity

Asia

Brazil

US/Canada

Australia

Europe

Asia: 344Brazil: 297US/Canada: 1201Australia: 294Europe: 656

Page 22: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

Frame Rate vs. User Geographic Location

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 5 10 15 20 25 30

Average Frame Rate (fps)

Cu

mm

ula

tive

De

ns

ity

Australia/Newzealand

US/Canada

Asia

Europe

Page 23: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

Frame Rate vs. PC Type

Page 24: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

Outline

• Introduction

• RealVideo

• Methodology

• Results

• Analysis– Frame Rate– Jitter – Perceived Quality

• Conclusions

Page 25: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

Jitter

• Frame rate alone not enough– Also need smooth playout– High frame rate but choppy playout can be as

bad as frame loss [CT99]

• Variation in frame playout interval called jitter– Standard deviation

• Buffering can reduce jitter [RKTS94, SJ95]

– But can still have residual jitter

• Quantitative impact of residual jitter on perceived quality scarce– 50 ms imperceptible, 300 ms very rough

Page 26: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

Overall Jitter

0102030405060708090

100

50 550 1050 1550 2050 2550 3050

Jitter (ms)

Cu

mu

lati

ve

De

ns

ity (%

)

Page 27: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

Jitter vs. Network Configuration

0

10

20

30

40

50

60

70

80

90

100

50 550 1050 1550 2050 2550 3050Jitter (ms)

Cu

mu

lati

ve D

en

sit

y (%

)

56K ModemDSL/CableT1/LAN

Page 28: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

Jitter vs. Server Geographic Location

0

10

20

30

40

50

60

70

80

90

100

50 1050 2050 3050

Jitter (ms)

Cu

mu

lati

ve

De

ns

ity (

%)

AustraliaUS/Canada

AsiaEuropeBrazil

Page 29: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

Jitter vs. User Geographic Location

0

10

20

30

40

50

60

70

80

90

100

50 550 1050 1550 2050 2550 3050

Jitter (ms)

Cum

ulat

ive

Dens

ity (%

)

Australia/New zealand

US/Canada

Asia

Europe

Page 30: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

Jitter vs. Observed Bandwidth

0

10

20

30

40

50

60

70

80

90

100

50 550 1050 1550 2050 2550 3050

Jitter (ms)

Cu

mu

lati

ve D

en

sit

y (%

)

< 10K

10K - 100K

> 100K

Page 31: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

Outline

• Introduction

• RealVideo

• Methodology

• Results

• Analysis– Frame Rate – Jitter– Perceived Quality

• Conclusions

Page 32: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

Perceived Quality

• Frame Rate and Jitter not enough– Best frame rate depends upon content [TC01]

• RealServer varies

– Effects of jitter depend upon content [CT99]

• Perceptual Quality– Numeric rating, 0-10

Page 33: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

00.10.20.30.40.50.60.70.80.9

1

0 5 10 15 20 25 30 35

Clips Rated per User

Cu

mu

lati

ve D

en

sit

y

(388 video clips rated)

Clips Rated per User

Page 34: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

Quality vs. Network Configuration

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 1 2 3 4 5 6 7 8 9 10

Quality Rating

Cu

mu

lati

ve D

en

sity

LAN/T1

DSL/Cable

56k Modem

Page 35: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

Conclusions

• Average RealVideo clip has good quality– 10 frames per second

– Nearly imperceptible jitter

• Network configuration– Modems are a bottleneck

– DSL/Cable modems pushing bottleneck towards server

Page 36: An Empirical Study of RealVideo Performance Across the Internet

November 2001 SIGCOMM Internet Measurement Workshop

Future Work

• RealVideo over UDPTCP-Compatible?

• Perceptual quality of streaming video

• Other streaming players and protocols– MediaPlayer, QuickTime

• Effects of caching, CDNs

• Release of customizable RealTracer

• Download IP addresses, Playlist, RealTracer– From Web page

Page 37: An Empirical Study of RealVideo Performance Across the Internet

An Empirical Study of RealVideo Performance Across the Internet

Yubing Wang, Mark Claypool and Zheng Zuo

http://perform.wpi.edu/real-tracer