thesis : "ibbet : in band bandwidth estimation for lan"

46
IBBET : In Band Bandwidth Estimation for LAN Vishalkumar Soni Thesis Advisor : Dr. Yusuf Ozturk MS Electrical & Computer Engineering Department of Electrical & Computer Engineering Computer Networks Research Lab Spring 2012

Upload: vishalkumarec

Post on 27-May-2015

404 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

IBBET : In Band Bandwidth Estimation for LAN

Vishalkumar SoniThesis Advisor : Dr. Yusuf Ozturk

MS Electrical & Computer EngineeringDepartment of Electrical & Computer Engineering

Computer Networks Research LabSpring 2012

Page 2: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

OutlineOutline

Need for Bandwidth Estimation Bandwidth Metrics Network Pipe Model Types of Bandwidth Estimation Packet Dispersion In Band Bandwidth Estimation Hypothesis : IBBET Pearson Correlation Coefficient Regression Analysis Matlab Implementation & Results IBBET Implementation Flowchart of Server Flowchart of Client Test Bed Setup IBBET Test Results Conclusion

Page 3: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

Need for Bandwidth EstimationNeed for Bandwidth Estimation

Rate based Streaming Application Verification of Quality of Service (QOS) Routing of packets Admission Control Resource Management

Page 4: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

Bandwidth MetricsBandwidth Metrics

Bandwidth : The maximum number of bits a link can transfer per unit of time [6].

Consider a network between two computer separated by n hops Narrow Link : Link with minimum capacity sets upper bound for

capacity of whole path

(1)

Available Bandwidth : The unused capacity of link for given interval of time

: Available Bandwidth

: Link with minimum capacity along the path

: Measured Cross traffic Tight Link : The link with minimum available bandwidth along the

network path

)()( tRCtAB xnw

}min{ in CC ni .....2,1

)(tABw

nC

)(tRx

(2)

Page 5: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

Network Pipe ModelNetwork Pipe Model

Consider LAN network as pipe model [6].

Processing Delay : The time to process packet through protocol stack

Latency : The time spend by packet during transmission in a link Queuing Delay : The time spent by packet in FIFO buffer of

router due to cross traffic

)( QLPD

D: Total delay experienced by the packet route from sender to receiverP: Processing delay experienced by packetL: Latency delay experienced by packetQ: Queuing delay experienced by packet

(3)

Page 6: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

Types of Bandwidth EstimationTypes of Bandwidth Estimation

Active Probing Probe packets are sent to estimate bandwidth Principle : If transmission rate of packets exceeds available

bandwidth then it increases queuing delay and reduce reception rate

e.g. Train of Packet Pair [6], PathChirp [3]

Passive Probing Application packets are monitor to estimate bandwidth Principle : It is based on round trip time of acknowledgement packet

for corresponding TCP packet sent

e.g. Passive Access Capacity Estimation [4], Idle Gap [8]

Page 7: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

Packet Dispersion Packet Dispersion

Packet Dispersion technique over three link model [1].

Capacity of link : C and Packet Size : L Consider two packets are send back-to-back as shown above Transmission delay : Receiver measures capacity of link as :

CL

LC

Page 8: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

In Band Bandwidth EstimationIn Band Bandwidth Estimation

Major Classification of Bandwidth Estimation

Out of band In band

Cons for using out of band bandwidth estimation technique

Congestion Latency Degrades overall utilization of channel Degrades QOS for real time application such as audio or video

Page 9: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

Hyothesis : IBBETHyothesis : IBBET

Reshaping application packets such as audio or video to infer bandwidth

Principle : Packet Dispersion Traffic shaper at server

◦ Reshapes stream of application packets

◦ Parameters : Packet size, Inter departure time Characteristics of traffic shaper

◦ Reshape application packets in a network friendly manner

◦ Minimum number of application packets in train Transmitted signature pattern is distorted due to bandwidth

limitation of network The receiver needs to detect signature of transmitted pattern from

distorted pattern using one of auto correlation function Regression analysis is performed to infer bandwidth

Page 10: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

Pearson Correlation CoefficientPearson Correlation Coefficient

It is defined as covariance of two variables divided by product of standard deviation [10].

(4)

For paired data of n data samples, the sample Pearson correlation coefficient is

(5)

: Sample mean

: Sample standard deviation

Correlation Coefficient Interpretation Negative Positive

None -0.09 to 0.0 0.09 to 0.0

Small -0.3 to -0.1 0.3 to 0.1

Medium -0.5 to -0.3 0.3 to 0.5

Large -1.0 to -0.5 0.5 to 1.0

yxyx

YX

),cov(

,

)( , ii YX

)])([(11

1

y

mi

x

min

i

YYXX

nr

mX mY

x y

Page 11: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

Regression Analysis Regression Analysis It is used for modeling relationship between dependent variable and one or

more independent variables [10]. The mathematical regression model can be represented as follows

: Dependent variable

: Independent variable

: Unknown parameters A linear regression model is represented as

(7) For linear regression, the unknown parameters are computed as

: Mean of values

: Mean of values

),( XfY Y

X

ii xy 10 ˆˆ

))((

))((ˆ1xxxx

yyxx

ii

ii xy 10 ̂

x

y

x

y

(8)

(6)

Page 12: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

Matlab ImplementationMatlab Implementation

Algorithm1. Specify configuration of signature pattern and network constrain.

2. Generate signature pattern for defined number of iterations.

3. The signature pattern is distorted based on user emulation of bottleneck link capacity and network distortion.

4. At receiver, transmitted signature pattern is recognized from distorted signature pattern using pearson correlation coefficient.

5. Linear regression analysis is applied to fit line between reception rate and time stamps of distorted signature pattern packets.

6. It estimates the slope and intercept for six consecutive packets.

7. The algorithm estimates bandwidth per pattern as average of reception rate of packets with slope less than threshold

8. The bandwidth for pattern stream is calculated as average of bandwidth per pattern.

Page 13: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

Matlab ImplementationMatlab Implementation

The following are condition applied to signature pattern and network

  Enter maximum rate of the transmitted probe signal in Mbps: 15 Enter constant rate of the stream in Mbps: 3 Enter number of packets for constant rate of stream: 12 Enter number of iteration of cycle: 4 Enter bottleneck link capacity: 9 Enter amount of network distortion in percentage: 6 Enter size of the packet in bytes: 1024

The estimated bandwidth at receiver is 8.8083 Mbps

Page 14: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

Matlab ImplementationMatlab Implementation

Transmitted Signature Pattern

Page 15: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

Matlab ImplementationMatlab Implementation

Received Signature Pattern

Page 16: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

Matlab ImplementationMatlab Implementation

Pattern detection

Page 17: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

IBBET ImplementationIBBET Implementation

Signature Pattern Equations

ntftfWhere )1()(

)()1()( ttftfWhere

)()1()( ttftfWhere

)(ty : Inter departure time of packets

: Constant exponent coefficient

T : Initial constant inter departure time of packets

)(tT : Time varying initial inter departure time of packets

n : Constant increment

)(t : User defined time varying increment

Where,

(9)

(10)

(11)

Page 18: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

IBBET ImplementationIBBET Implementation

0 Mbps 100 Mbps

10

25

70

90

Search Window

Bandwidth Range

Signature Pattern Shape

Page 19: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

Flow Chart of ServerFlow Chart of ServerStart

Define Destination Port & IP Address

Configure Probing Pattern

Create UDP Socket

Sock == -1

Exit

Assign Destination IP & Port to Server Address

Structure

Memory Allocation To

Buffer

Initialization of Packet

NO

Current Iteration <= DefineCycles

CurrentPacket <= Defined Packets of

Probe Train

Send Packet

Estimate Inter Departure Time

Log Transmission Rate,Inter Departure Time & Packet Sequence

Change Sleep period based on probing equation

YES

YES

EndNO

Send Packets for Constant Stream

Sleep for Constant Period

CurrentPacket <= Defined Constant Stream Packets

NO

NO

YES

Page 20: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

Flow Chart of ClientFlow Chart of ClientReceive Packet

Estimate Inter Arrival Time

Estimate Reception Rate per Packet

Logging of Reception Rate per packet, inter arrival time, Packet Sequence

End of Probing Stream

Read expected inter arrival time of probe stream

CurrentPattern<= Defined Pattern

Measure difference between expected & actual inter arrival time of Packets

Error < Thresold

Store Supported Receive Rate

End of Pattern Packets

Pattern Bw = Last Supported Received Rate *Scale Factor

YES

YES

YES

YES

NO

NO

NO

NO

Bw = Avg. Pattern Bw over No. of Received Pattern

End

Start

Define Port

Create UDP Socket

Sock == -1 Exit

Assign Port & IP Address to Client Address Structure

Bind Socket to Listening Port

bind == -1 Exit

NO

YES

YES

NO

A

A

Page 21: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

Test Bed SetupTest Bed Setup

Server : Streams signature pattern WANem : Puts bandwidth constrain on interface as per configuration Client : Estimates the bandwidth

Router

ClientWANem

Emulator PCServer

Physical Connection

Flow of Packets

Page 22: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

IBBET ResultsIBBET Results

Number of packets in signature pattern : 40 Incremental : 0.2 Alpha coefficient : 0.7 Packet size : 1024 B Probing range : 1 to 18 Mbps Constant rate stream : 3 Mbps Number of constant rate stream packets : 15 WANem bandwidth constrain : 4 Mbps Number of iterations : 5

Tty tf )()( ntftfWhere )1()(

Configuration of Signature Pattern & WANem

Signature Pattern from equation (9)

Page 23: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

IBBET ResultsIBBET Results

Inter departure time between packets at server for equation (9)

Page 24: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

IBBET ResultsIBBET Results

Inter arrival time between packets at client for equation (9)

Page 25: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

IBBET ResultsIBBET Results

Bandwidth (Mbps) Pattern

Sequence

3.368421 1

4.535991 56

3.521926 111

4.818823 166

4.108325 221

Avg. Bandwidth = 3.368421 + 4.535991+ 3.521926+4.818823+ 4.108325 5

Avg. Bandwidth = 4.070697 Mbps

Bandwidth estimation for equation (9)

Page 26: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

IBBET ResultsIBBET Results

Number of packets in Signature Pattern : 40 Incremental : 0.2 Alpha Coefficient : 0.7 Packet Size : 1024 B Probing Range : 1 to 18 Mbps Constant Rate Stream : 3 Mbps Number of Constant Rate Stream packets : 15 WANem bandwidth Constrain : 5 Mbps Number of Iterations : 15

ntftfWhere )1()(

Configuration of Signature Pattern & WANem

The standard deviation is computed based on following equation

1

)~

(1

2

N

XXiSD

N

i

Signature Pattern from equation (9)

Page 27: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

IBBET ResultsIBBET Results

Bandwidth estimation for equation (9)

Page 28: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

IBBET ResultsIBBET Results

Standard deviation for equation (9)

Page 29: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

IBBET ResultsIBBET Results

Number of packets in Signature Pattern : 40 Incremental value for first 30 Packet : 0.2 Incremental value for remaining 10 Packet : 0.4 Alpha Coefficient : 0.7 Packet Size : 1024 B Probing Range : 1 to 28 Mbps Constant Rate Stream : 3 Mbps Number of Constant Rate Stream packets : 15 WANem bandwidth Constrain : 5 Mbps Number of Iterations : 5

Configuration of Signature Pattern & WANem

)()1()( ttftfWhere

Signature Pattern from equation (10)

Page 30: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

IBBET ResultsIBBET Results

Inter departure time between packets at server for equation (10)

Page 31: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

IBBET ResultsIBBET Results

Inter arrival time between packets at client for equation (10)

Page 32: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

IBBET ResultsIBBET Results

Bandwidth (Mbps) Pattern

Sequence

6.989761 1

4.452174 56

6.400000 111

4.830189 166

3.416180 221

Bandwidth estimation for equation (10)

The average bandwidth over the five iteration is 5.217660 Mbps

Page 33: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

IBBET ResultsIBBET Results

Number of packets in Signature Pattern : 40 Incremental value for first 30 Packet : 0.2 Incremental value for remaining 10 Packet : 0.4 Alpha Coefficient : 0.7 Packet Size : 1024 B Probing Range : 1 to 28 Mbps Constant Rate Stream : 3 Mbps Number of Constant Rate Stream packets : 15 WANem bandwidth Constrain : 5 Mbps Number of Iterations : 15

Configuration of Signature Pattern & WANem

Signature Pattern from equation (10)

)()1()( ttftfWhere

The standard deviation is computed based on following equation

1

)~

(1

2

N

XXiSD

N

i

Page 34: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

IBBET ResultsIBBET Results

Bandwidth estimation for equation (10)

Page 35: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

IBBET ResultsIBBET Results

Standard deviation for equation (10)

Page 36: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

IBBET ResultsIBBET Results

Number of packets in Signature Pattern : 40 Initial Probing Rate : 4 Mbps Time varying Incremental value for Packets : 0.1 Alpha Coefficient : 0.7 Packet Size : 1024 B Probing Range : 4 to 18 Mbps Constant Rate Stream : 3 Mbps Number of Constant Rate Stream packets : 15 WANem bandwidth Constrain : 6 Mbps Number of Iterations : 5

Configuration of Signature Pattern & WANem

)()( )( tTty tf )()1()( ttftfWhere

Signature Pattern from equation (11)

Page 37: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

IBBET ResultsIBBET Results

Inter departure time between packets at server for equation (11)

Page 38: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

IBBET ResultsIBBET Results

Inter arrival time between packets at client for equation (11)

Page 39: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

IBBET ResultsIBBET Results

Bandwidth (Mbps)Pattern

Sequence

5.535135 1

6.291859 56

6.671010 111

6.872483 166

6.095238 221

Bandwidth estimation for equation (11)

The average bandwidth over the five iteration is 6.293145 Mbps

Page 40: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

IBBET ResultsIBBET Results

Number of packets in Signature Pattern : 40 Initial Probing Rate : 3 Mbps Time varying Incremental value for Packets : 0.1 Alpha Coefficient : 0.7 Packet Size : 1024 B Probing Range : 3 to 13 Mbps Constant Rate Stream : 3 Mbps Number of Constant Rate Stream packets : 15 WANem bandwidth Constrain : 5 Mbps Number of Iterations : 15

Configuration of Signature Pattern & WANem

)()( )( tTty tf )()1()( ttftfWhere

Signature Pattern from equation (11)

The standard deviation is computed based on following equation

1

)~

(1

2

N

XXiSD

N

i

Page 41: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

IBBET ResultsIBBET Results

Bandwidth estimation for equation (11)

Page 42: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

IBBET ResultsIBBET Results

Standard deviation for equation (11)

Page 43: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

ConclusionConclusion

On comparing equations (9), (10) and (11) for WANem bandwidth constrain of 5 Mbps. Equation (11) has least standard deviation over period of 15 patterns.

Hence, we can say that equation (11) will require minimum number of probes to get fairly accurate bandwidth estimation.

Moreover, equation (11) provides more parameters to change in order to adapt signature pattern to dynamically changing bandwidth.

Our Matlab simulation and WANem test results revels that IBBET estimate bandwidth fairly accurate.

This algorithm will reduces congestion on network and reduces time for rate adaptation at server.

Page 44: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

ReferencesReferences[1] Constantious Dovrolis, Parameswaram Ramanathan and David More,“What do a packet dispersion technique

measure?”, INFOCOM, Twentieth Annual Joint Conference of the IEEE Computer and Communications Societies, IEEE, p.905 - 914 vol.2,2001.

[2] Bob Melander, Mats Bjorkman and Per Gunningberg, “A New End-to-End Probing and Analysis method for Estimating Bandwidth Bottlenecks”,Proc.IEEE, GLOBECOM, 2000.

[3] Vinay J. Ribeiro, Rudolf H. Riedi, Richard G. Baraniuk, Jiri Navratil, and Les Cottrell, “pathChirp: Efficient available bandwidth estimation for network paths”, In Passive and Active Measurment Workshop, April 2003.

[4] Fornasa, Martino, Maresca and Massimo, “Passive Access Capacity Estimation” ,International Journal of Network Management, NEM-09-0070, University of Padova, 2009. 

[5] Ningning Hu and Peter Steenkiste, “Estimating Available Bandwidth Using Packet Pair Probing”, School of Computer Science, Carnegie Mellon University, Pittsburg, PA, 2002.

[6] Dimas Lopez Villa and Carlos Ubeda Castellanos, “Study of Available Bandwidth Estimation Techniques to be applied in Packet-Switched Mobile Networks”, Department of Communication Technology, Aalborg University, 2006.

[7] Abhik Majumdar, Daniel Grobe Sachs, Igor V. Kozintsev, Kannan Ramchandra, and Minerva M. Yeung, “Multicast and Unicast Real-Time Video Streaming Over Wireless LANs”, IEEE Transactions on Circuits and Sysems for Video Technology, Vol. 12, No. 6, June 2002. 

[8] Heung Ki Lee, Varrian Hall, Ki Hwan Yum, Kyoung Ill Kim and Eun Jung Kim, “Bandwidth Estimation in Wireless LAN for multimedia streaming services”, Texas A & M University, University of Texas San Antanio, Electronics & Telecommunication Research Institute.

[9] Mingzhe Li, Mark Claypool and Robert Kinicki, “WBest : A Bandwidth Estimation tool for Multimedia Streaming Application for IEEE 802.11 Wireless Networks”, Computer Science Department at Worcester Polytechnic Institute, Worchester, MA, USA.

[10] Walter A. Rosenkrantz “Introduction to probability and statistics for scientist and engineers” McGraw Hill Series In, ISBN 0-07-053988-X.

Page 45: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

Thank YouThank You

Page 46: Thesis : "IBBET : In Band Bandwidth Estimation for LAN"

Questions ??Questions ??