18
Advanced Network Management (Fall 2007) Professor: James Won-Ki Hong Chen Bin Kuo (20077202) Young J. Won (20063292)

ppt

Embed Size (px)

Citation preview

Page 1: ppt

Advanced Network Management (Fall 2007)

Professor: James Won-Ki Hong

Chen Bin Kuo (20077202)

Young J. Won (20063292)

Page 2: ppt

DPNM Lab.

Revisited

VoIP, Data, & Other Traffic Models

TPS Models

IPTV Live Traffic Model

IPTV VoD Traffic Model

Conclusion

04/13/23 2

Page 3: ppt

DPNM Lab. 04/13/23 3

Details

Bundled Services Traffic Selection & Survey on Previous Models

Measurement Analysis on Selected Bundled Service Applications

Measurement Paper (IPTV)

Model Formulation

Traffic Impact & Demand Analysis

Final Report

Page 4: ppt

DPNM Lab.

Problem?Lack of effective combinational models for analyzing traffic impact and demand in bundled services environment

Traffic demand analysis in bundled services Traffic impact analysis with existing traffic

Goal? Traffic monitoring and analysis of bundled service applications Traffic modeling formulation for combined traffic

04/13/23 4

Page 5: ppt

DPNM Lab.

Link performance summary

04/13/23 5

Capture summary

Page 6: ppt

DPNM Lab.

The D&P type completes its download with initial traffic bursts The streaming type show low yielding bandwidth through

the entire running time

04/13/23 6

Page 7: ppt

DPNM Lab.

Asymmetric traffic delivery pattern 60~90 bytes in size for signaling purpose to acknowledge the stuffed (1500

bytes) video packets Upload performance shows no impact on the overall quality of IPTV STB,

unlike P2P IPTV

04/13/23 7

Page 8: ppt

DPNM Lab.

Traffic bandwidth fluctuation Watching two consecutive episodes 7 Mbps vs. 2 Kbps No channel surfing traffic burst, unlike multicast model

This implies that there could be much simpler representation of IPTV traffic models

04/13/23 8

Page 9: ppt

DPNM Lab.

Measurement conditions Location: San Francisco, USA Broadband access: Cable STB: MegaTV Viewing duration: 30 minutes / 120 minutes

Quick observations Buffering delay in about every 5 minutes

Occasional frame stoppage Below minimum throughput bound (3.5 Mbps vs. 6 Mbps) Fluctuation vs. Constant

0 5 10 15 20 251.0

1.5

2.0

2.5

3.0

3.5

4.0 Download Throughput - MegaTV over Cable, USA

Time (Minutes)

Th

ro

ug

hp

ut

(Mb

ps

)

Down

04/13/23 9

Page 10: ppt

DPNM Lab.

The bandwidth at the client domain does not correspond to the required SD or HD transfer ratios

The proposed formulas describe the IPTV VoD services by D&P delivery architecture which has not been proposed in any other work

The traffic burst due to channel surfing is negligible in this VoD architecture

The running time of each channel is fixed and known The occurrence of VoD traffic is not continuous but an independent discrete

event

Channel viewing time does not necessarily coincide with the packet transmission time between the server and STB

04/13/23 10

Page 11: ppt

DPNM Lab.

Initialization(viewers, program files)Initialization(viewers, program files)

Choose viewers randomly(number of viewers is Poisson distribution)

Viewers choose the program file (Zipf distribution)

Calculate bandwidth demand from active viewers

Check viewers states(downloading, playing, finishing)

Simulation ends(time is up)Simulation ends(time is up)

1 round = 1 minute1 round = 1 minute

04/13/23 11

Page 12: ppt

DPNM Lab.

Parameter Description

Maximum active viewers 200

Mena number of requests by viewers 5 (request/minute)

Total Simulation Duration 200 minutes

04/13/23 12

File size (Mbytes) 2000 1500 1000 500 250 200 200 200 200 200

Playing Duration (Minutes)

120 80 60 30 15 10 10 10 10 10

Popularity 11.38% 17% 34.1% 8.53% 6.82% 5.7% 4.88% 4.27% 3.8% 3.4%

Access Network FTTB FTTH xDSL Cable

Downloading Rate(Mbps)

10 11 6 7

Popularity 20% 10% 60% 10%

• Assumptions• Viewer behavior is uniform• Once the viewer chose the program file, the

viewer stayed until the program file ends• # of requests per unit time by viewers

follows Poisson distribution• Popularity of program files follow Zipf

distribution• These can be more realistic if we have the

real values.

Page 13: ppt

DPNM Lab. 04/13/23 13

( ) ( ) ( )j jj

B t D t t (1)

, if ( )

0, otherwise

l jk jk kl

j

r T t TD t

(2)

kjk

l

f

r (3)

FTTH

FTTBl

Cable

xDSL

r

rr

r

r

(4)

jk jk kC T o

(5)

1ll

p

(6)

1kk

p

(7)

k

cp

k

(8)

,( )

,,

( )!

tm

tt

eP m

m

(9)

,[ ]tE m

(10)

Notation Description

kf File size of program file k, where 1,2,3...k

lr Downing rate by media l, where

, , ,l FTTH FTTB Cable xDSL

lp Popularity of medium l

kp Popularity of program file k

,tm Number of requests by viewers during time period

,t t

Mean number of requests by viewers per unit time

jkT The time that viewer j started to download program file k

kl The duration of downing program file k by medium l

ko Playing duration of program file k

jkC The time that viewer j stopped to play program k (channel change)

( )jD t The bandwidth demand of user j at time t

( )j t A function of t, which is 1 if viewer j is on at time t, otherwise 0.

( )B t Total bandwidth demand of an aggregate link at time t

Page 14: ppt

DPNM Lab.

Multicast Live TV simulation vs. VoD simulation Avg. 1000 Mbps vs. Avg. 550 Mbps for 200 viewers Over 2 hours (multiple channel viewings)

Take the average of multiple simulations Test 1-4 and its average

04/13/23 14

Page 15: ppt

DPNM Lab.

TPS Components

VoIP VoIP+Data Data HTTP IPTV Others

Available Choices

1

5

6

1

2

2 2 3

4

3+4

6

Number Methodologies

1. Market projection based

2. Peak-to-average analysis

3. Multicast Live Demand Model

4. VoD Demand Model

5. VoIP Demand Model (Erlang-B,C)

6. Constant Rate

04/13/23 15

Page 16: ppt

DPNM Lab.

VoIP+Data, IPTV Market-based approach, Multicast demand model Market-based approach, VoD demand model Market-based approach, Multi + VoD demand model

VoIP, Data, IPTV Call model, Peak-to-average analysis, Demand model

VoIP, HTTP, UDP, IPTV Call model, Peak-to-average analysis, Constant, Demand model

From two perspectives Edge, Backbone Minimum and Maximum bound Data-dominated IPTV-dominated

04/13/23 16

Page 17: ppt

DPNM Lab.

Contributions A guideline on what to look for and how to handle TPS or

beyond traffic demand analysis A concrete and available set of models are explained

1) Previous VoIP, Data, IPTV, and TPS as a whole for bandwidth demand models

2) Our proposed models for IPTV and simulation3) Combinational model examples

04/13/23 17

Page 18: ppt