Network Simulation

1
Network Simulation

• Dave Lin
• 2001/3/31
• R89055_at_im.ntu.edu.tw

2
Overview

• Simulation Modeling
• Probability and Statistics in Simulation
• Elements of Discrete Simulation
• Queue Model Example
• MAC and PHY Layer Simulation Example
• Conclusion
• References

3
Simulation Modeling
4
Why Simulation?

• To represent a real system by a model.
• Simulation is the next best ting to observing a
  real system in operation.
• To collect pertinent observations about the
  system over time.
• Queuing theory? Queuing models are rather
  restrictive.

5
Types of Simulation

• Discrete
• Observation are gathered only at selected points
  in time when certain changes take place in the
  system.
• Ex. M/M/1 model
• Continuous
• Observations be collected continuously at every
  point in time.
• Ex. Oil level in tank

6
How the Simulation Clock Ticks

• Automatic internal clock is designed to initiate
  the action of collecting observations at the
  moment changes take palace in the system.
• Continuous simulation
• Looking at the system at equally spaced time
  intervals.

7
How the Simulation Clock Ticks (Contd)

• Discrete Simulation
• System is observed only when specific changes
  occur.
• Ex. single source, single queue, single server.

A1
A2
A3
A4
Simulation Time
D1
D2
8
Randomness Element in Simulation

• Process of arrival at and departure from a
  facility occurs in a random fashion.
• Only way we can represent this randomness is by
  using probability distributions to describe the
  variables under consideration.
• Ex. Poisson arrival process
• Ex. Random select between two servers

9
Discrete Simulation Languages

• Event-scheduling
• Deal directly with the individual action
  associated with the occurrence of each event
  type.
• Highly flexible in representing complex situation
  since user must provide details.
• Process-based
• Represent the movement of an instance from it
  enters the system until it is discharged.
• Relieving the user form programming most of the
  detial.

10
Other Simulation Language

• Object-Oriented Language
• Natural Simulation Language
• Good and easy way to model system
• Network Simulation Language
• OPNET Modeler - down to finite state machine,
  state logic
• COMNET III network nodes, links and protocols

11
Two Sides of Simulation

• Analysis, Design, Construction, Testing and
  running of the model
• Interpretation of model output
• Most crucial and probably the most neglected
  among simulation users.
• Simulation study must be subjected to all the
  proper statistical techniques just as in any
  other physical experiment.

12
Probability and Statistics in Simulation
13
Role of Probability and Statistics in Simulation

• Randomness is a key element in the description of
  the system.
• Simulation output must be viewed as a sample in a
  statistical inference tests.
• Measures of performance of a simulation model
  typically must be expressed in terms of
  appropriate confidence intervals.

14
Identifying Distribution Based on Historical Data

• Summarizing historic or observed data in the form
  of an appropriate histogram.
• Hypothesizing a theoretical distribution based on
  the shape of the histogram and then testing its
  goodness-of-fit using an appropriate statistical
  model.
• Chi-Square Test
• Kolmogrov-Smirnov Test

15
Statistical Output Analysis

• Confidence Intervals
• Simulation results should thus be presented in
  the form of statistical confidence intervals.
• Hypothesis Testing
• Hypothesis testing is used in simulation
  experiments when it is desired to test whether a
  mean of a given measure of performance is equal
  to a specified value.

16
Elements of Discrete Simulation
17
Concept of Events in Simulation

• In discrete simulation, changes in the systems
  statistics are triggered by the occurrences of
  certain key event.
• The use of events is fundamental to the
  development of all discrete simulation languages.

18
Common Simulation Approaches

• Event-Scheduling Approach
• Activity-Scanning Approach
• Process-Simulation Approach

19
Event-Scheduling Approach

• Event include two basic attributes
• Type of event
• Occurrence time
• Event calendar file (E.FILE)
• List which store event chronologically
• Represents the automatic time advancing mechanism

20
Event-Scheduling Approach (Contd)

• Single source, single queue, single server
  example
• Gather information (statistics)
• Length of QUEUE
• Waiting time in QUEUE
• Utilization of SERVER
• Idle time of FACILITY
• Events
• Arrival Event
• Departure Event

21
Event-Scheduling Approach (Contd)

• Arrival Event
• Generate and store chronologically the occurrence
  time of the next arrival ( Current simulation
  time interarrival time)
• If SERVER is idle
• Enter customer in service and indicate the SERVER
  is busy
• Update SERVER idle time statistics.
• Generate and store chronologically the occurrence
  time of the departure event ( current simulation
  time service time)
• Go to 3c.
• If SERVER is busy
• Place the customer in QUEUE.
• Update QUEUE length statistics.
• Destroy the current event, select the next one on
  time scale.

22
Event-Scheduling Approach (Contd)

• Departure Event
• IF QUEUE is empty
• Declare the server idle.
• Go to 2c.
• If QUEUE is not empty
• Remove a customer from queue and update QUEUE
  waiting time statistics.
• Generate and store the occurrence time of the
  departure event ( current simulation time
  service time).
• Destroy the current event and select the next one
  on the time scale.

23
Event-Scheduling Approach (Contd)
Select the most imminent event (next event on
the time scale)
Implement departure event actions and insert new
events chronologically in the E.FILE
Implement arrival event actions and insert new
events chronologically in the E.FILE
Y
N
Arrival Event?
24
Activity-Scanning Approach

• Similar to the event-scheduling approach.
• Events
• Arrival event
• Departure event
• Service start event
• Arrival Event
• Place arrival customer in QUEUE.
• Generate and sore the next arrival event.

25
Activity-Scanning Approach (Contd)

• Departure Event
• Declare SERVER idle.
• Service Start Event
• Remove a customer from QUEUE and start service in
  SERVER.
• Generate departure event from SERVER.
• Service start event would be executed under two
  conditions are satisfied
• QUEUE is not empty.
• SERVER is idle.

26
Activity-Scanning Approach (Contd)

• Check to execute service start event after each
  Arrival event or Departure event is executed
  repeatedly.
• Obviously inefficient process of repeatedly
  checking.
• Usually is simpler than event scheduling.
• Lack of widespread use in practice.

27
Process-Simulation Approach

• Same concept of events utilized in event
  scheduling.
• Processor of simulation language with no
  interference form the user, deals with
  transaction generation, statistics collection
  jobs.
• More compact than in event scheduling.
• Less flexibility than event scheduling approach.
• Needs simulation language support.

28
Collection Data in Simulation

• Observation-based and Timer-based statistical
  variables.
• Histograms.
• Queue statistics
• Avg. queue length, avg. waiting time
• Server statistics
• Avg. utilization, avg. busy time per server, avg.
  idle time per server.

29
Queue Model Example

• M/M/1 example
• M/M/m/m example

30
Queue Model Example
Timer
Server 1
. . .
Recorder
Traffic Generator
Traffic Queue
Server m
31
MAC and PHY Layer Simulation Example

• Bluetooth MAC layer
• 802.11b MAC layer
• Coexistence of Bluetooth and 802.11b PHY layer

32
BTG
WTG
Timer
BMACM
WMAC SEND
BMACS
WMAC REC
PHY
BR
WR
PECM
33
Conclusion

• Special attributes of wireless network simulation
• Time domain and Frequency domain interference.
• Simulation is a useful tool for studying network
  behaviors.
• There are good rules and tools for the
  construction network simulation.

34
References

• Hamdi A. Taha, Prentice-Hall ,Simulation Modeling
  and Simnet ISBM 0-13-8811621-0