Network Design and Analysis-----Wang Wenjie Introduction: 1 © Graduate University, Chinese academy of Sciences. Network Design and Performance Analysis.

Network Design and Analysis-----Wang Wenjie Introduction: 1

© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Network Design and

Performance Analysis

Wang Wenjie

[email protected]


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

What the Course is about

• 随着计算机通信、电信网络等系统的发展，统计性能评价方法变得越来越重要。统计性能评价是关于复杂的网络或软件等系统的行为预测的问题，它是使用数学方法以及计算机等工具进行分析的方法。

• This course is about computer communications network planning , design and analysis


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Learning Objectives(1)

• Upon successful completion of this course, you will be able to …

– Apply analytical models to evaluate the relative performance of single and multiple queue systems used to model networks and protocols

– Describe traffic models used for packet switched traffic and discuss the concept of self-similarity and its applicability to network modeling

– Design simulation experiments, develop traffic models, develop network models, and analyze results for evaluating the performance of queues, local area networks, internetworks, and protocols


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Learning Objectives(2)

• Upon successful completion of this course, you will be able to (continued) …

– Apply modeling and network design techniques to design data networks based on performance objectives and design constraints

– Explain the use of network management tools and network monitors to determine network performance and operational problems


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Course Agenda

绪论随机变量和随机过程简介排队论自相似过程 Petri网络介绍多路访问协议拥塞控制路由选择


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Course Information

• Who is the course for?

– Graduate Students

• Grading

– Homework:• Assignments • Programming

– Final Exam :


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Reading list

• Bertsekas D. & Gallager R., Data Networks – Second Edition, Prentice-Hall, 1992

（中文：数据网络，人民邮电出版社）• Stallings W., High-Speed Networks: TCP/IP and ATM

Design Principles, Second edition, Prentice Hall, 2002

（中文：高速网络与互联网，电子工业出版社）• Thomas G. Robertazzi, Computer Networks and

Systems: Queueing Theory and Performance Evaluation, 3rd edition, Springer-Verlag, 2000 (ISBN 0-387-95037-0).


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Introduction


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Outlines

• Network Models

• Methods for Performance Evaluation

• Quality of Service(QoS)

• Traffic Definitions


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Outlines

• Network Models





© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Network Models

• What are we modeling?

• Networks (broadly defined) generally fall into one of three categories

– Resource sharing networks– Process networks– Switching networks


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Resource sharing networks(1)

• Time-shared computer (Programs : CPU/Disk/IO)

• Statistical multiplexer/concentrator

• Packet-based (Packets : Links)

• Channel-based (Calls : Channels)

• Multiple-Access and random access networks (Packets : Shared Medium)


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Resource sharing networks(2)

• Performance measures

– Waiting time– Blocking probability

• Questions

– What is the relationship between the number of users, the pattern of usage, the amount of resources, and the performance?

– Are resources utilized fairly (or appropriately) by the users?


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Process networks(1)


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Process networks(2)


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Switching networks(1)


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Switching networks(2)


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Outlines

• Network Models





© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Life Cycle

Performance Evaluation/Analysis

Network Design

Network Management


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Design

• A Network Design is a blueprint for building a network. The Designer has to create the structure of the network [and] decide how to allocate resources and spend money

( R. Cahn)

• Tools: Capacity Assignment techniques , graph theory, computer projects


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Management

• “(…) our only real option is to use the network to manage the network. This means we need a protocol that allows us to read, and possibly write, various pieces of state information on different nodes.”

(L. Peterson & B. Davie)

• Tools: SNMP, experiments


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Performance Evaluation

• We will predict the behavior of networks “using pencil, paper, the poetry of mathematics and the number-crunching ability of computers.”

(T. Robertazzi)

• Tools: Queuing theory, traffic models, simulation


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Need for Performance Evaluation

• The most important concern with a network system is that it correctly perform its intended functions

• The second concern is that it has “adequate” performance at a “reasonable” cost

– “adequate performance” / “reasonable cost” may be explicitly or implicitly given, and usually determined largely by the intended functions

– In general, performance and cost are given little consideration


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Methods for Performance Evaluation

There are three main methods used in the design of performance studies

• Experimental approaches

– measurement and use of a real system

• Analytic approaches

– the use of mathematics, queueing theory, Petri Nets, abstract models, etc

• Simulation approaches

– design and use of computer simulations and simplified models to assess performance


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Example: ISP(1)


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Example: ISP(2)


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Example: ISP(3)


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Example: ISP(4)


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Example: ISP(5)


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Example: ISP(6)


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Example: ISP(7)

• Verification (usually easy to do)• Is the model properly constructed?• Analytical model: “Is the math right?”• Simulation model: “Is the model properly coded?”

• Validation (often hard to do)• Does the model accurately represent the behavior of the system of interest?• Analytical model: “Is the model ‘correct’?”• Simulation model: “Is this an accurate simulation for the questions of interest?”


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Outlines

• Network Models





© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Quality of Service(1)

• Variety of definitions exist in literature

• ATM definition- “Quality of Service is the performance observed by an end user”

• Internet definition - Still evolving :-)

• QoS is also usually expressed as the combination of network-imposed delay, jitter, bandwidth, loss and reliability----------refers to providing a level of service “acceptable” to network users, e.g.,

– Blocking rate: Percent of calls blocked that user can accept

– Delay:Delay time should be less than y milliseconds

– Jitter:Cell delay variation


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Quality of Service(2)

Important Performance ParametersSystem BlockingDelay in the messagingJitterBuffer OverflowPacket LossSystem ReliabilityEnd-to-end delay in media packetsResponse DelaySystem Recovery TimeAdmission Control AlgorithmResource Scheduling algorithmsSystem CapacitySystem Engineering Rules

Simulation &Queuing Theory

Capacity

Game Theory


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Attributes of QoS(1)

• QoS defines a set of attributes related to the performance of connection.

– User-related attributes: the attributes that define how fast the user wants to send data.

– Network-related attributes: the attributes are those that define characteristics of the network

• QoS parameters need to be mapped between layers

• Application layer QoS Frame rate, size of video

• Network layer QoS bandwidth, Delay


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Attributes of QoS(2)

• Peceptual Parameter System QoS

• Picture detail Pixel resolution

• Picture color accuracy Maps to color information per pixel

• Video Rate Maps to frame rate

• Video smoothness Maps to frame rate jitter

• Audio Quality Audio Sampling rate and number of bits

• Video/audio synchronisation Video and audio stream synchronised

• for example lip-sync.


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Why needs QoS?(1)


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Why needs QoS?(2)

• The level of service is one of the key parameters to determine how much resources should be put in the network

• Equipment utilization:

– This is an associated issue with QoS/GoS , it is about proper utilization of resources


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

What need for QoS?(3)

• With the notion of QoS for a certain service, we also need:

– Minimally, have some information on the amount of demand (which typically refers to as offered traffic in the context of computer communication networks)

– Have a way to character the estimate of traffic


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Outlines

• Network Models





© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Traffic and Traffic Demand

• A network provider has control over design and management for the network that is under its own administrative domain; thus, in order to do that , an important need for each provider would be to determine traffic demand in its own network

• Traffic in the Internet

• Traffic in the Telephone Network


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Traffic in the Internet

• Applications: emails, ftp, web, …

• We do not know ahead of time when and where to access, use or whether to request a web page ?

--- The Internet data traffic has a statistical property

• Measurement systems need to capture statistics so that traffic arrival distribution and traffic volume can be estimated.

--- M/M/1,…


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Traffic in the Telephone Network

• The voice circuit is dedicated for one user.

---- Average call arrival rate

---- Average duration of a call

• offered load or offered traffic is telephony

product of the two factors

• Other notions: Erlang-B loss, carried load or traffic


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Example of traffic system （ 1 ）

• There are three nodes

• Suppose:

1)Traffic given in the number of connected is deterministic

2)Each connection requires one unit of capacity for a direct link, two units of capacity through a third node

3)Demand units for a link is direction-less


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Example of traffic system （ 2 ）

Morning(M) Afternoon(A) Evening(E)

AB 5 11 8

AC 6 13 6

BC 10 2 7

A C

B


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Example Analysis

• If we know how to design the network with capacity and routing rule in place, and

If it is properly utilized,

then the demand can be met

• Network planning involves determination:

Network topologyCapacity links and switchesRouting


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Second problem

• Suppose that we have determined the capacity and put them in place , and now the traffic changed slightly, the question is what is the maximum number of connections that can accommodated given the capacity?


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Third problem

• Consider arrival of two packets streams to a network node. If we have some kind of flow control activated for one stream, how is the system behavior for these traffic streams?


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.

Traffic Definition(1)

QoS Framework

• Static Functions– Traffic and QoS specifications (traffic types/parameters)– QoS negotiation and signalling– Admission control– Resource reservation

• Dynamic Functions– Traffic shaping and policing– Queuing and scheduling – Congestion control


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.


Traffic Source Types

• CBR (Constant Bit Rate) : transmits traffic at a fixed rate, such as 64 Kbps voice

• VBR (Variable Bit Rate) : traffic rate is not fixed; sometimes high, sometimes low, such as MPEG coded video


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.


Traffic Parameters

• Different flows have different traffic patterns

• A given traffic pattern can be described using several traffic parameters

• Peak rate : maximum rate in any time interval

• Average rate : long term average

• Burst size : duration of peaks


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.


peak rate

average rateburst size

Time

bps


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.


• All patterns have the same average rate (10 Kbps), but different peak rate and burst size

time

100 Kbps

50 Kbps

10 Kbps


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.


• In before example:

Traffic is the number of connections

• In general , the traffic offered to a group of devices (lines, circuits, links , trunks, or traffic paths) can be specified by two parameters:

– The average arrival rate – The average service time , the reciprocal of is

referred to as the service rate .


© G

rad

uat

e U

niv

ersi

ty ,

Ch

ines

e ac

adem

y o

f S

cien

ces.


• For data networks:

Offered traffic=avg number of packets/time

• For voice(telephone) networks:

Offered load =

(Number of call attempts/hour)*avg call holding time

• This unit is Erlang

One erlang represents a circuit occupied for one hour

• Carried traffic = Offered traffic – lost traffic

Network Design and Analysis-----Wang Wenjie Introduction: 1 © Graduate University, Chinese academy of Sciences. Network Design and Performance Analysis.

Documents