1

Introduction To Simulation

A Simulation is the imitation of the operation of a real world process or a facility. The
facility or the process of interest is usually called a system.

In order to study the system we make a set of assumptions about it. These assumptions
constitute a model. Assumptions are expressed in mathematical or logical relationship.

If the model is simple enough, it may be solved by mathematical methods such as

calculus, algebra or probability theory. However, many real world problems are too
complex. Models of these systems are almost impossible to solve analytically. In a
simulation we use a computer to evaluate a model numerically and data are gathered in
order to estimate the desired characteristics of the model.

Systems and System Environment

To model a system, it is necessary to understand the concept of a system and the

system boundary. A system is defined as a group of objects that are joined together in
some regular interaction or interdependence toward the accomplishment of some
purpose.

A system is often affected by changes occurring outside the system. Such changes are
said to occur in the system environment. In modeling systems, it is necessary to decide
on the boundary between the system and its environment. This decision may depend on
the purpose of the study.

Component of a System

In order to understand and analyze a system, a number of terms need to be defined. An

entity is an object of interest in the system. An attribute is a property of an entity. An
activity represents a time period of specific length. If a bank is being studied, customers
might be one of the entities, the balance in their accounts might be an attribute, and
making deposits might be an activity.

The state of a system is defined to be that collection of variables necessary to describe

the system at any time, relative to the objectives of the study. In the study of a bank,
possible state variables are the number of busy tellers, the number of customers waiting
in line or being served, and the arrival time of the next customer. An event is defined as
an instantaneous occurrence that may change the state of the system. The term
 2

endogenous is used to describe activities and events occurring within a system and the
term endogenous is used to describe activities and events occurring within a system and
the term exogenous is used to describe activities and events in the environment that
affect the system. In the bank study, the arrival of a customer is an
endogenous event, and the completion of service of a customer is an endogenous event.

Discrete and Continuous Systems

System can be categorized as discrete or continuous. A discrete system is one in which

the state variable(s) change only at a discrete set of points in time. The bank is an
example of a discrete system, since the state variable, the number of customers in the
bank, changes only when a customer arrives or when the service provided to a customer
is completed.

A continuous system is one in which the state variables(s) change continuously over
time.

Model of a System

A model is defined as a representation of a system for the purpose of studying the

system. For most studies, it is necessary to consider only those aspects of the system
that affect the problem under investigation. These aspects are represented in a model of
the system, and the model, by definition, is a simplification of the system. On the other
hand, the model should be sufficiently detailed to permit valid conclusions to be drawn
about the real system. Different models of the same system may be required as the
purpose of investigation changes.

Types of Models

Models can be classified as being mathematical or physical. A mathematical model uses

symbolic notation and mathematical equations to represent a system. A simulation model
is a particular type of mathematical model of system.

Simulation models may be further classified as being static or dynamic, deterministic or

stochastic, and discrete or continuous. A static simulation model, sometimes called a
Monte Carlo simulation, represents a system at a particular point in time. Dynamic
simulation models represent systems as they change over time. The simulation of a bank
from 9:00 A.M. to 4:00 P.M. is an example of a dynamic simulation.

Simulation models that contain no random variables are classified as deterministic.

Deterministic model have a known set of inputs, which will result in a unique set of
outputs.
 3

Steps in Simulation Study

Problem formulation: Every study should begin with a statement of the problem. If the
statement is provided by the policy makers or those that have the problem, the analyst
must ensure that the problem being described is clearly understood.

Setting of objectives and overall project plan

The objective indicates the questions to be answered by simulation. At this point a

determination should be made concerning whether simulation is the appropriate
methodology for the problem as formulated and objective as stated. The overall project
plan should include the plans for the study in terms of the number of people involved, the
cost of the study and the number of days required to accomplish each phase of the work
with the anticipated results at the end of each stage.

Model conceptualization

The construction of a model of a system is probably as much art as science. The art of
modeling is enhanced by an ability to abstract the essential features of a problem. To
select and modify basic assumptions that characterize the system, and then to enrich and
elaborate the model until a useful approximation results, the model complexity need not
exceed that required to accomplish the purposes for which the model is intended.
Violation of this principle will only add to model-building expenses.

It is advisable to involve the model user in model conceptualization. This will both
enhance the quality of the resulting model and increase the confidence of the model user
in the application of the model.

Data collection

There is a constant interplay between the construction of the model and the collection of
the needed input data. As the complexity of the model changes, the required data
elements may also change.

Model translation

Since most real-world systems result in models that require a great deal of information
storage and computation, the model must e entered into a computer-recognizable format.
We use the term “program”, even though it is possible to accomplish the desired result in
many instance with little or no actual coding. The modeler must decide whether to
program the model in a simulation language or to use special-purpose simulation
software.
 4

Verifications

Verification pertains to the computer program prepared for the simulation model. Is the
computer program performing properly? If the input parameters and logical structure of
the model are correctly represented in the computer, verification has been completed.

Validation

Validation is the determination that a model is an accurate representation of the real

system. Validation is usually achieved through the calibration of the model, an iterative
process of comparing the model to actual system behavior and using the discrepancies
between the two, and the insights gained to improve the model. This process is repeated
until model accuracy is judged acceptable.

Experimental design

The alternatives that are to be simulated must be determined. Often, the decision
concerning which alternatives to simulate may be function of runs that have been
completed and analyzed.

Production runs and analysis

Production runs and their subsequent analysis, are used to estimate measures of
performance for the system designs that are being simulated.

Documentation and reporting

There are two types of documentation: program and progress, Program documentation is
necessary for numerous reasons. If the program is going to be used again by the same
or different analysts, it may be necessary to understand how the program operates.

Implementation

The success of the implementation phase depends on how well the previous eleven
steps have been performed. It is also contingent upon how thoroughly the analyst has
involved the ultimate model user during the entire simulation process.

Advantages and Disadvantages of Simulation:

Advantages are

1. New policies, operating procedures, decision rules, information flows,

organizational procedures, and so on can be explored without disrupting ongoing
operations of the real system.
 5

2. New hardware designs, physical layouts, transportation systems, and so on, can
be tested without committing resources for their acquisition.

3. Hypotheses about how or why certain phenomena occur can be tested for
feasibility.

4. Time can be compressed or expanded allowing for a speedup or slowdown of the

phenomena under investigation.

5. Insight can be obtained about the interaction of variables.

6. Insight can be obtained about the importance of variables to the performance of

the system.

7. Bottleneck analysis can be performed indicating where work-in-process

information. Materials, so on are being excessively delayed.

8. A

9. “What-if”, questions can be answered. This is particularly useful in the design of

new systems.

Disadvantages are

1. Model building requires special training. It is an art that is learned over time and
through experience. Furthermore, if two models are constructed by two
competent individuals, they may have similarities, but it is highly unlikely that they
will be the same.

2. Simulation results may be difficult to interpret. Since most simulation outputs are
essentially random variables, it may be hard to determine whether an
observation is a result of system interrelationships or randomness.

3. Simulation modeling and analysis can be time consuming and expensive.

Skimping on resources for modeling and analysis may result in simulation model
or analysis that is not sufficient for the task.

4. Simulation is used in some cases when an analytical solution is possible, or even

preferable.

When simulation is the appropriate tool

Simulation can be used for the following purposes.

 6

1. Simulation enables the study of the internal interactions of a complex system or a

subsystem within a complex system.

2. Informational, organizational and environmental changes can be simulated and

their effect on the model can be observed.

3. By changing simulation input and observing the resulting outputs, we may know
which variables are important.

4. Simulation can be used to experiment with new designs or policies so as to

estimate its effect.

5. Simulation can be used to verify analytic solution.

6. Simulation models designed for training allow learning without the cost and
efforts of on the job training.

7. Using animation we can visualize the system in a particular stage of operation.

8. Many modern systems can be so complex that simulation may be the only way to
study them.

When simulation is not appropriate?

Simulation should not be used.

1. If the problem can be solved analytically.

2. If the problem can be solved using common sense.

3. If it is easier to perform direct experiment.

4. If the costs exceed the savings.

5. If the resources and time are not available.

6. If proper and sufficient data are not available.

7. If there is no time or skilled personnel are not available to verify and validate the
model.

8. If system behavior is extremely complex or cannot be defined.

9. If there is an unreasonable expectation.

 7

Graded Questions

1. What is system modeling? Give an example and explain the different types of
models. [N-04]

2. Define simulation. When is simulation appropriate and when it is not? [N-04]

3. Define simulation. What are the various steps in simulation study? [M-05]

4. Define model. What are the different types of models? Give example for each.
[M-05]


 8

2
Simulation Examples

Graded Questions

1. A small grocery store has only one checkout counter. Customers arrive at this
checkout counter at random from 1 to 8 minutes apart. Each possible value of
interarrival time has the same probability of occurrence. The service time distribution
is as follows:

Service time (minutes) : 1 2 3 4 5 6

Probability : 0.05 0.10 0.20 0.30 0.25 0.10

Perform the simulation for 10 customers. Obtain average waiting time in the queue
and average idle time of the server.

2. Observations of past data show the following patterns in respect of inter-arrival

durations and service durations in a single channel queuing system. Using the
random number table below, simulate the queue behaviour for a period of 60 minutes
and estimate the probability of the service being idle and the mean time spent by a
customer waiting for service.

Inter-arrival Time Service Time

Minutes Probability Minutes Probability
2 0.15 1 0.10
4 0.23 3 0.22
6 0.35 5 0.35
8 0.17 7 0.23
10 0.10 9 0.10

Random numbers (start at North-West corner and proceed along the row)

93 14 72 10 21
81 87 90 38 10
29 17 11 68 99
51 40 30 52 71

3. A plant has a large number of similar machines. The machine breakdowns or failures
are random and independent.
 9

The shift in-charge of the plant collected the data about the various
machines breakdown times and the repair time required on hourly basis, and
the record for the past 100 observations as shown below was:

Time Between Probability Repair Time Probability

Recorded Required
Machine (hours)
Breakdown
(hours)
0.5 0.05 1 0.28
1 0.06 2 0.52
1.5 0.16 3 0.20
2 0.33
2.5 0.21
3 0.19

For each hour that one machine is down due to being or waiting to be repaired, the
plant loses Rs. 70 by way of lost production. A repairman is paid at Rs. 20 per hour.
a. Simulate this maintenance system for 15 breakdowns.
b. Obtain the total maintenance cost.

Use following pairs of random numbers:

(61,87), (85,39),(16,28),(46,97),(88,69),(08,87),(82,52),(56,52),(22,15),(49,85)

4. The management of a company is considering the problem of marketing a new

product. The investment or the fixed cost required in the project is Rs. 25,000. There
are three factors that are uncertain-selling price, variable cost and the annual sales
volume. The product has a life of only one year. The management has the past data
regarding the possible levels of the three factors.

Unit Selling Probability Unit Probability Sales Probability

Price Variable Volume
(Rs.) Cost (units)
(Rs.)
40 0.30 20 0.10 3,000 0.20
50 0.50 30 0.60 4,000 0.40
60 0.20 40 0.30 5,000 0.40

Using MonteCarlo simulation technique, determine the average profit from the said
investment on the basis of 20 trials.
 10

5. A company manufactures 200 motorcycles per day. Depending upon the availability
of raw materials and other conditions, the daily production has been varying from 196
motorcycles to 204 motorcycies, whose probability distribution is as given below:

Production/day: 196 197 198 199 200 201 202 203 204
Probability: 0.05 0.09 0.12 0.14 0.20 0.15 0.11 0.08 0.06

The motorcycles are transported in a specially designed three-storeyed lorry

that can accommodate only 200 motorcycles. Using the following random

numbers:

82,89,78,24,52,53,61,18,45,04,23,50,77,27,54,10, simulate the process to find out:

a. The average number of motorcycles waiting in the factory?
b. The average number of empty spaces on the lorry?

6. A confectioner sells confectionary items. Past data of demand per week (in hundred
kilograms) with frequency is given below:
7.
Demand/week: 0 5 10 15 20 25
Frequency: 2 11 8 21 5 3

Using the following sequence of random numbers, generate the demand for the next
10 weeks. Also find the average demand per week:
35,52,90,13,23,73,34,57,35,83,94,56,67,66,60

8. Consider the following continuously operating job shop. Interarrival times of jobs are
distributed as follows:
Time between Probability
Arrivals (Hours)
0 .23
1 .37
2 .28
3 .12

Processing times for jobs are normally distributed with mean 50 minutes and
standard deviation 8 minutes. Construct a simulation table, and perform a simulation
for 10 new customers. Assume that when the simulation beings there is one job
 11

being processed (scheduled to be completed in 25 minutes) and there is one job with
a 50-minute processing time in the queue.
(a) What was the average time in the queue for the 10 new jobs?
(b) What was the average processing time of the 10 new jobs?
(c) What was the maximum time in the system for the 10 new jobs?

9. A plumbing supply firm is interested in the distribution of lead-time demand of

industrial sinks. The probability distribution for daily for daily demand is known and
occurs as show:

Daily 0 1 2 3 4
Demand
Probability 0.18 0.39 0.29 0.09 0.05

The distribution of lead-time has been reconstructed from past records as

follows:

Lead Time 0 1 2 3 4 5
(Days)
Probability 0.135 0.223 0.288 0.213 0.118 0.023

Develop the distribution of lead-time demand based on 20 cycles of lead-time.

10. Given A, B, C, these independent random variables: Variable A is normally

distributed with  = 100 and  = 400. Variable B is discrete uniformly distributed with
2

probability distribution given by p (b) = 1/5 with b = 0,1,2,3 and 4. Variable C is

distributed in accordance with the following table.

Value of C Probability
10 .10
20 .25
30 .50
40 .15

Use simulation to estimate the mean of a new variable D, defined as

D = (A – 25B) / (2C)

Use a sample of size 10.

 12

11. Lead-time for a stock item is normally distributed with a mean of 7 days and standard
deviation of 2 days. Daily demand is distributed as follows:

Daily 0 1 2 3 4
Demand
Probability 0.367 0.368 0.184 0.062 0.019

Determine the lead-time demand for 20 order cycles. (Round off lead time to the
closest integer during the simulation, and if a negative value results, give it a lead
time of zero.)

12. A bank has one drive-in teller and room for one additional customer to wait.
Customers arriving when the queue is full, park and go inside the bank to transact
business. The time-between-arrivals and services-time distributions are given below.

Time between Probability Service Time Probability

Arrivals (Minutes) (Minutes)
0 0.09 1 0.20
1 0.17 2 0.40
2 0.27 3 0.28
3 0.20 4 0.12
4 0.15
5 0.12

Simulate the operation of the drive-in-teller for 10 new customers. The first of the 10 new
customers arrives at a time determined at random. Start the simulation with one customer
being served, leaving at time 3, and one in the queue. How many customers went into
the bank to transact business?
Table Random Normal Numbers

0.23 -0.17 0.43 2.18 2.13 0.49 2.72 -0.18 0.42

0.24 -1.17 0.025 0.67 -0.59 -0.13 -0.15 -0.46 1.64
-1.16 -017 0.36 -1.26 0.91 0.71 -1.00 -1.09 -0.02
-0.02 -0.19 -0.04 1.92 0.71 -0.90 -0.21 -1.40 -0.38
0.39 0.55 0.13 2.55 -0.33 0.05 -0.34 -1.95 -0.44
0.64 -0.36 0.98 -0.21 -0.52 -0.02 -0.15 -0.43 0.62
-1.90 0.48 -0.54 0.60 -0.35 -1.29 -0.57 0.23 1.41
-1.04 -0.70 -1.69 1.76 0.47 -0.52 -0.73 0.94 -1.63
-7.8 0.11 -0.91 -1.13 0.07 0.45 -0.94 1.42 0.75
0.68 1.77 -0.82 -1.68 -2.60 1.59 -0.72 -0.80 0.61
-0.02 0.92 1.76 -0.66 0.18 -1.32 1.26 0.61 0.83
-0.47 1.04 0.83 -2.05 1.00 -0.70 1.12 0.82 0.08
 13

-0.40 1.40 1.20 0.00 0.21 -2.13 -0.22 1.79 0.87

-0.75 0.09 -1.50 0.14 -2.99 -0.41 -0.99 -0.70 0.51
-0.66 -1.97 0.15 -1.16 -0.60 0.50 1.36 1.94 0.11
-0.44 -0.09 -0.59 1.37 0.18 1.44 -0.80 2.11 -1.37
1.41 -2.71 -0.67 1.83 0.97 0.06 -0.28 0.04 -0.21
1.21 -0.52 -0.20 -0.88 -0.78 0.84 -1.08 -0.25 0.17
0.07 0.66 -0.51 -0.04 -0.84 0.04 1.60 0.92 1.14
-0.08 -0.09 -1.12 0.40 0.69 -0.12
0.53 -0.36 -2.64 0.22 -0.78 1.92 -0.26 1.04 -1.61
-1.56 1.82 -1.03 1.14 -0.12 -0.78 -0.12 1.42 -0.52
0.03 -1.29 -0.33 2.60 -0.64 1.19 -0.13 0.91 0.78
1.49 1.55 -0.79 1.37 0.97 0.17 0.58 1.43 -1.29
-1.19 1.35 0.16 1.03 -0.17 0.32 -0.28 0.68 0.54
-1.19 -1.03 -0.12 1.07 0.87 -1.40 -0.24 -0.81 0.31
0.11 -1.95 -0.44 -0.39 -0.15 -1.20 -1.98 0.32 2.91
-1.86 0.06 0.19 -1.29 0.33 1.51 -0.36 -0.80 -0.99
0.16 0.28 0.60 -0.78 0.67 0.13 -0.47 -0.18 -0.89
1.21 -1.19 -0.60 -1.22 0.07 -1.13 1.45 0.94 0.54
-0.82 0.54 -0.98 -0.13 1.52 0.77 0.95 -0.84 2.40
0.75 -0.80 -0.28 1.77 -0.16 -0.33 2.43 -1.11 1.63
0.42 0.31 1.56 0.56 0.64 -0.78 0.04 1.34 -0.01
-150 -1.78 -0.59 0.16 0.36 1.89 -1.19 0.53 -0.97
-0.89 0.08 0.95 -0.73 1.25 -1.04 -0.47 -0.68 -0.87
0.19 0.85 1.68 -0.57 0.37 -0.48 -0.17 2.36 -0.53
0.49 0.32 -2.08 -1.02 2.59 -0.53 0.15 0.11 0.05
-1.44 0.07 -0.22 -0.93 -1.40 0.54 -1.28 -0.15 0.67
-0.21 -0.48 1.21 0.67 -1.10 -0.75 -0.37 0.68 -0.02
-0.65 -0.12 0.94 -0.44 -1.21 -0.06 -1.28 -1.51 1.39
0.24 -0.83 1.55 0.33 -0.59 -1.24 0.70 0.01 0.15
-0.73 1.24 0.40 -0.61 0.68 0.69 0.07 -0.23 -0.66
-1.93 0.75 -0.32 0.95 1.35 1.51 -0.88 -0.10 -1.19
0.08 0.16 0.38 -0.96 1.99 -0.20 0.98 0.16 0.26
-0.47 -1.250 0.32 0.51 -1.04 0.97 2.60 -0.08 1.19

Simulation Examples
Chap – 2

Simulation of Queuing Systems –

 14

A queuing system is described by its calling population, the nature of the arrivals, the
service mechanism, the system capacity & the queuing discipline.

Calling population waiting line server

Fig. Queuing System

In the single channel queue, the calling population is infinite ie if a unit leaves the
calling population & joins the waiting lines or entire service, there is no change in the
arrival rate of other units that may need service.

In the single queue, the system works an FIFO by a single server or channel.

Arrivals & services are defined by the distribution by the distribution of service
times, resp. for any simple single or multi channel queue the overall effective motival sale
must be lets than the total service rate or the waiting line wall grow without bound. When
queues grow without bound they are termed explosive or unstable.

1. The average waiting time for a customer 2.8 minutes. This is determined in the
following manner.

Average waiting time total time customers wait in queue (minutes)

=
(minutes) total numbers of customers
56
= = 2.8 minutes
20

2. The probability that a customer has to wait in the queue is 0.65. This is
determined in the following manner:
Number of customers who wait
Probability (wait) =
Total number of customers

13
= = 0.65
20
 15

3. The fraction of idle time of the server is 0.21. This is determined in the following
manner:
Probability of idle total idle tune if server (minutes)
Server =
Total run time of simulation (minutes)
18
= 86 = 0.21
The probability of the server being busy is the complement of 0.21, or 0.79.

4. The average service time is 3.4 minutes, determined as follows:

Average service time total service time (minutes)
(minutes) =
total number of customers
68
= = 3.4 minutes
20

This result can be compared with the expected service time by finding the mean of the
service time by finding the mean of the service-time distribution using the equation


E (S) =  s p (s)
S=0
Applying the expected-value equation to the distribution in Table 2.7 gives an expected
service time of:
= 1(0.10) + 2 (0.20) + 3 (0.30) + 4 (0.25) + 5 (0.10) + 6 (0.05)
= 3.2 minutes

the expected service time is slightly lower that the average service time in the simulation.
The longer the simulation, the closer the average will be to E (S).

5. The average time between arrivals is 4.3 minutes. This is determined in the
following manner:
Sum of all times
Average time between = between arrivals (minutes)
Arrivals (minutes)
Number of arrivals – 1
82
= = 4.3 minutes
19
One is subtracted from the denominator because the first arrival is assumed to occur at
time 0. This result can be compared to the expected time between arrivals by finding the
 16

mean of the discrete uniform distribution whose endpoints are a = 1 and b = 8. The mean
is given by
a+b 1+8
E (A) = = 4.5 minutes
2 2
The expected time between arrivals is slightly higher than the average. However, as the
simulation becomes longer, the average value of the time between arrivals will approach
the theoretical mean, E (A).

6. The average waiting time of those who wait is 4.3 minutes. This is determined in
the following manner.
Average waiting time of total time customers wait in queue (minutes)
=
those who wait (minutes) total number of customers who wait
56
= = 4.3 minutes
13

7. The average time a customer spends in the system is 6.2 minutes. This can be
determined in two ways. First, the computation can be achieved by the following
relationship:
Total time customers spend in the
Average time customer system (minutes)
Spends in the system =
(minutes) total number of customers

124
= = 6.2 minutes
20

The second way of computing this same result is to realize that the following relationship
must hold:
Average time average time average time
Customers spends customer spends customer spends
In the system = waiting in the + in service
(minutes) queue (minutes) (minutes)

From finding 1 and 4 this results in:

Average time customer spends in the system (minutes)

= 2.8 + 3.4 = 6.2 minutes

 17

A decision maker would be interested in results of this type, but a longer

simulation would increase the accuracy of the findings. However, some subjective
inferences can be drawn at this point. Most customers have to wait; however, the
average waiting time is not excessive. The server does not have an undue amount of idle
time. Objective statements about the results would depend on balancing the cost of
waiting with the cost of additional servers. (Simulations requiring variations of the arrival
and service distribution, as well as implementation in qa spreadsheet, are presented as
exercises for the reader.)

2.2 Simulation of Inventory Systems

An important class of simulation problems involves inventory systems. A simple

inventory system is shown in Figure 2.7. This inventory system has a periodic review of
length N, at which time the inventory level is checked. An order is made to bring the
inventory up to the level M. At the end of the first review period, an order quantity, Q1, is
placed. In this inventory system the lead-time (i.e., the length of time between the
placement and receipt of an order) is zero. Since demands are not usually known with
certainty, the order quantities are probabilistic. Demand is shown as being uniform over
the time period in Figure and do fluctuate over time. One possibility is that demands all
occur at the beginning of the cycle. Another is that the lead time is random of some
positive length.

Figure 2.7 Probabilistic order-level inventory systems.

Notice that in the second cycle, the amount in inventory drops below zero,
indicating a shortage. In Figure 2.7, these units are backordered; when the order arrives,
the demand for the backordered items is satisfied first. To avoid shortages, a buffer, or
safety, stock would need to be carried.
 18

Carrying stock in inventory has an associated cost attributed to the interest paid
on the funds borrowed to buy the items (this also could be considered as the loss from
not having the funds available for other investment purposes). Other costs can be placed
in the carrying or holding cost column: renting of storage space, hiring guards, and so on.
An alternative to carrying high inventory is to make more frequent reviews, and
consequently, more frequent purchases or replenishments. This has an associated cost:
the ordering cost. Also, there is a cost in being short. Customers may get angry, with a
subsequent loss of good will. Larger inventories decrease the possibilities of shortages.
These costs must be traded off in order to minimize the total cost of an inventory system.

The total cost (or total profit) of an inventory system is the measure of
performance. This can be affected by the policy alternatives. For example, in Figure 2.7,
the decision maker can control the maximum inventory level, M, and the length of the
cycle, N. What effect does changing N have on the various costs?

In an (M, N) inventory system, the events that may ovvur are: the demand for
items in the inventory, the review of the inventory position, and the receipt of an order at
the end of each review period. When the lead-time is zero, as in Figure 2.7, the last two
events occur simultaneously.

In the following example for deciding how many newspapers to buy, only a single
time period of specified length is relevant and only a single procurement is made.
Inventory remaining at the end of the single time period is sold for scrap or discarded. A
wide variety of real-world problems are of this form, including the stocking of spare parts,
perishable items, style goods, and special seasonal items [Hadley and Whitin, 1963].

SUMMARY:

This chapter introduced simulation concepts via eg in order to illustrate general areas of
application.

Ad hoc simulation tables were used in completing each eg.,


 19

3
General Principles

3.1 Concepts in Discrete-Event Simulation

The concept of a system and a model of a system were discussed briefly in

Chapter1. This chapter deals exclusively with dynamic, stochastic systems (i.e., involving
time and containing random elements) with change in a discrete manner. This section
expands on these concepts and develops a framework for the development of a discrete-
event model of a system. The major concepts are briefly defined and then illustrated by
examples:

System A collection of entities (e.g., people and machines) that interact together
over time to accomplish one or more goals.

Model An abstract representation of a system, usually containing structural, logical,

or mathematical relationships which describe a system in terms of state, entities and
their attributers, sets, processes, events, activities, and delays.

System state A collection of variables that contain all the information necessary to
describe the system at any time.

Entity Any object or component in the system, which requires explicit representation
in the model (e.g., a server, a customer, a machine).

Attributes The properties of a given entity (e.g., the priority of a waiting customer,
the routing of a job through a job shop).

List A collection of (permanently or temporarily) associated entities, ordered in some

logical fashion (such as all customers currently in a waiting line, ordered by first
come, first served, or by priority).

Event An instantaneous occurrence that changes the state of a system (such as an

arrival of a new customer).

Event notice A record of an event to occur at the current or some future time, along
with any associated data necessary to execute the event; at a minimum, the record
includes the event type and the event time.
 20

Event list A list of event notices for future events, ordered by time of occurrence;
also known as the future even list (FEL).

Activity A duration of time of specified length (e.g., a service time or interarrival

time), which is known when it begins (although it may be defined in terms of a
statistical distribution).

Delay A duration of time of unspecified indefinite length, which is not known until it
ends (e.g., a customer’s delay in a last-in, first-out waiting line which, when it begins,
depends on future arrivals).

Clock A variable representing simulated time, called CLOCK in the examples to

follow.

Different simulation packages use different terminology for the same or similar
concepts. For the same or similar concepts. For example, lists are sometimes called sets,
queues, or chains. Sets or lists are used to hold entities as well as event notices. The
entities on a list are always ordered by some rule, such as first-in, first-out or last-in, first-
out, or ranked by some entity attribute, such as priority or due date. The future event list
is always ranked by the event time recorded in the event notice. Section 3.2 discusses a
number of methods for handling lists and introduces some of the methodologies for
efficient processing of ordered sets or lists.

An activity typically represents a service time, an interarrival time, or any other

processing time whose duration has been characterized and defined by the modeler. An
activity’s duration may be specified in a umber of ways:

1. Deterministic – for example, always exactly 5 minutes;

2. Statistical - for example, as a random draw from among 2,5,7 with equal
probabilities;

3. A function depending on system variables and/or entity attributes – for example,

loading time for an iron ore ship as a function of the ship’s allowed cargo weight
and the loading rate in tons per hour.

However it is characterized, the duration of an activity is computable from its

specification at the instant it begins. Its duration is not affected by the occurrence of other
events (unless, as is allowed by some simulation packages, the model contains logic to
cancel an event). To keep track of activities and their expected completion time, at the
simulated instant that an activity duration begins, an event notice is created having an
event time equal to the activity’s completion time. For example if the current simulated
time is CLOCK = 100 minutes and an inspection time of exactly 5 minutes is just
 21

beginning, then an event notice is created that specifies the type of event (an end of
inspection event) and the event time (100 + 5 = 105 minutes).

In contrast to an activity, a delay’s duration is not specified by the modeler ahead

of time, but rather is determined by system conditions. Quite often, a delay’s duration is
measured and is one of the desired outputs of a model run. Typically, a delay ends when
some set of logical conditions becomes true or one or more other events occur. For
example, a customer’s delay in a waiting line may be dependent on the number and
duration of service of other customers ahead in line as well as the availability of servers
and equipment.

A delay is sometimes called a conditional wait, while an activity is called an

unconditional wait. The completion of an activity is an event, often called a primary event,
that is managed by placing an event notice on the FEL. In contrast, delays are managed
by placing the associated entity on another list, perhaps representing a waiting line, until
such time as system conditions permit the processing of the entity. The completion of a
delay is sometimes called a Conditional or Secondary Event, but such events are not
represented by event notices nor do they appear on the FEL.

A discrete-event simulation is the modeling over time of a system all of whose

state changes occur at discrete points in time-those points when an event occurs. A
discrete-event simulation (hereafter called a simulation) proceeds by producing a
sequence of system snapshots (or system images) which represent the evolution of the
system through time. A given snapshot at a given time (CLOCK = t) includes not only the
system state at time t, but also a list (the FEL) of all activities currently in progress and
when each such activity will end, the status of all entities and current membership of all
sets, plus the current values of cumulative statistics and counters that will be used to
calculate summary statistics at the end of the simulation. A prototype system snapshot is
shown in Figure 3.1. (Not all models will contain every element exhibited in Figure 3.1.
Further illustrations are provided in the examples in this chapter.)

3.1.1 The Event-Scheduling/Time-Advance Algorithm

The mechanism for advancing simulation time and guaranteeing that all events occur in
correct chronological order is based on the future event list (FEL). This list contains all
event notices for events that have been scheduled to occur at a future time. Scheduling a
future event means that at the instant an activity begins, its duration is computed or
drawn as a sample from a statistical distribution and the end-activity event, together with
its event time, is placed on the future event list. In the real world, most future events are
not scheduled but merely happen-such as random breakdowns or random arrivals. In the
model, such random events are represented by the end of some activity, which in turn is
represented by a statistical distribution.

At any given time t, the FEL contains all previously scheduled future events and
their associated event times (called t1, t2, … in Figure 3.1). The FEL is ordered by event
 22

time, meaning that the events are arranged chronologically; that is, the event times
satisfy

T < t1 < t2 < t3 < . . .. < tn

Time t is the value of CLOCK, the current value of simulated time. The event
associated with time t1 is called the imminent event; that is, it is the next event that will
occur. After the system snapshot at simulation time CLOCK = t has been updated, the
CLOCK is advanced to simulation time CLOCK = t1, and the imminent event notice is
removed from the FEL and the event executed. Execution of the imminent event means
that a new system snapshot for time t1 is created based on the old snapshot at time t and
the nature of the imminent event. At time t1, new future events may or may not be
generated, but if any are, they are scheduled by creating event notices and putting them
in their proper position on the FEL. After the new system snapshot for time has been
updated, the clock is advanced to the time of the new imminent event and that event is
executed. This process repeats until the simulation is

The length and contents of the FEL are constantly changing as the simulation
progresses, and thus its efficient management in a computerized simulation will have a
major impact on the efficiency of the computer program representing the model. The
management of a list is called list processing. The major event, addition of a new event to
the list, and occasionally removal of some event (called cancellation of an event). As the
imminent event is usually at the top of the list, its removal is as efficient as possible.
Addition of a new event (and cancellation an old event) requires a search of the list. The
efficiency of this search depends on the logical organization of the list and on how the
search is conducted. In addition to the FEL, all the sets in a model are maintained in
some logical order, and the operations of addition and removal of entities from the set
also require efficient list-processing techniques. A brief introduction to list processing in
simulation is given in Section 3.2.

We can the removal and addition of events from the FEL is illustrated in Figure
3.2. Event 3 with event time t1 represents, say, a service-completion event at server 3.
Since it is the imminent event at time t, it is removed from the FEL in step 1 (Figure 3.2)
of the event-scheduling/time-advance algorithm. When event 4 (say, an arrival event)
with event time t* is generated at step 4, one possible way to determine its correct
position on the FEL is to conduct a top-down search:
If t* < t2 place event 4 at the top of the FEL.
If t2 < t* < t3, place event 4 second on the list.
If t3 < t* < t4, place event 4 third on the list.
:
:
If tn < t*, place event 4 last on the list.

(In figure 3.2, it was assumed that t* was between t2 and t3.) Another way is to conduct a
bottom-up search. The least efficient way to maintain the FEL is to leave it as an
 23

unordered list (additions placed arbitrarily at the top or bottom), which would require at
step 1 of Figure 3.2 a complete search of the list for the imminent event before each
clock advance. (The imminent event is the event on the FEL with the lowest event time.)

The system snapshot at time 0 is defined by the initial conditions and the
generation of the so-called exogenous events. The specified initial conditions define the
system state at time 0. For example, in Figure 3.2, if t = 0, then the state (5,1,6) might
represent the initial number of customers at three different points in the system. An
exogenous event is a happening “outside the system” which impinges on the system. An
important example is an arrival to a queuing system. At time 0, the first arrival event is
generated and is scheduled on the FEL (meaning its event notice is placed on the FEL).
The interarrival time is an example of an activity. When the clock eventually is advanced
to the time of this first arrival, a second arrival event is generated. First, an interarrival
time is generated, a*; it is added to the current time, CLOCK = t; the resulting (future)
event time, t + a* = t*, is used to position the new arrival event

Figure 3.3. Generation of an external arrival stream by bootstrapping.

Notice on the FEL. This method of generating an external arrival stream, called
bootstrapping, provides one example of how future events are generated in step 4 of the
event-scheduling/time-advance algorithm. Bootstrapping is illustrated in Figure 3.3. The
first three interarrival times generated are 3.7, 0.4, and 3.3 time units. The end of an
interarrival interval is an example of a primary event.

A second example of how future events are generated (step 4 of Figure 3.2) is
provided by a service-completion event in a queueing simulation. When one customer
completes service, at current time CLOCK = t, if the next customer is present, then a new
service time, s*, will be generated for the next customer. The next service-completion
event will be scheduled to occur at future time t* = t + s* by placing onto the FEL a new
 24

event notice of type service completion with event time t*. In addition, a service-
completion event will be generated and scheduled at the time of an arrival event,
provided that, upon arrival, there is at least one idle server in the server group. A service
time is an example of an activity. Beginning service is a conditional event, because its
occurrence is triggered only on the condition that customer is present and a server is
free. Service completion is an example of a primary event. Note that a conditional event,
such as beginning service, is triggered by a primary event occurring and certain
conditions prevailing in the system. Only primary events appear on the FEL.

A third important example is the alternate generation of runtimes and downtimes

for a machine subject to breakdowns. At time 0, the first runtime will be generated and an
end-of-runtime event scheduled. Whenever an end-of-runtime event occurs, a downtime
will be generated and an end-of-downtime event scheduled on the FEL. When the
CLOCK is eventually advanced to the time of this end-of-downtime event, a runtime is
generated and an end-of-runtime event scheduled on the FEL. In this way, runtimes and
downtimes continually alternate throughout the simulation. A runtime and a downtime are
examples of activities, and end of runtime and end of downtime are primary events.

Every simulation must have a stopping event, here called E, which defines how
long the simulation will run. There are generally two ways to stop a simulation:
1. At time 0, schedule a stop simulation event at a specified future time TE. Thus,
before simulating, it is known that the simulation will run over the time interval [0,
TE]. Example: Simulate a job shop for TE = 40 hours.

2. Run length TE is determined by the simulation itself. Generally, TE is the time of

th
the 100 service completion at a certain service center. TE is the time of
breakdown of a complex system. TE is the time of disengagement or total kill
(whichever occurs first) in a combat simulation. TE is the time at which a
distribution center ships the last carton in a day’s orders.

In case 2, TE is not known ahead of time. Indeed, it may be one of the statistics of
primary interest to be produced by the simulation.

3.1.2 World Views

When using a simulation package or even when doing a manual simulation, a

modeler adopts a world view or orientation for developing a model. Those most
prevalent are the event-scheduling world view, as discussed in the previous section,
the process-interaction world view, and the activity-scanning world view. Even if a
particular package does not directly support one or more of the world views,
understanding the different approaches may suggest alternative ways to model a
given system.
To summarize the previous discussion, when using the event-scheduling
approach, a simulation analyst concentrates on events and their effect on system
 25

state. This world view will be illustrated by the manual simulations of Section 3.1.3
and the C++ simulation in Chapter 4.
When using a package that supports the process-interaction approach, a
simulation analyst thinks in terms of processes. The analyst defines the simulation
model in terms of entities or objects and their life cycle as they flow through the
system, demanding resources and queueing to wait for resources. More precisely, a
process is the life cycle of one entity. This life cycle consists of various events and
activities. Some activities may require the use of one or more resources whose
capacities are limited. These and other constraints cause processes to interact, the
simplest example being an entity forced to wait in a queue (on a list) because the
resource it needs is busy with another entity. The process-interaction approach is
popular because of its intitive appeal, and because the simulation packages that
implement it allow an analyst to describe the process flow in terms of high-level block
or network constructs, while the interaction among processes is handled
automatically.
In more precise terms, a process is a time-sequenced list of events, activities,
and delays, including demands for resources, that define the life cycle of one entity
as it moves through a system.

An example of a “customer process” is shown in Figure 3.4. In this figure, we

see the interaction between two customer processes as customer n + 1 is
delayed until the previous customer’s “end-service event” occurs. Usually
many processes are simultaneously active in amodel, and the interaction
among processes may be quite complex.
Underlying the implementation of the process-interaction approach in a
simulation package, but usually hidden from a modeler’s view, events are being
scheduled on a future event list and entities are being placed onto lists when-ever
they face delays, causing one process to temporarily suspend its execution while
other processes proceed. It is important that the modeler have a basic understanding
 26

of the built-in but hidden rules of operation. Schriber and Brunner [1998] provide
understanding in this area.
Both the event-scheduling and the process-interaction approaches use a variable
time advance; that is, when all events and system state changes have occurred at
one instant of simulated time, the simulation clock is advanced to the time of the next
imminent event on the FEL. The activity-scanning approach, in contrast, uses a fixed
time increment and a rule-based approach to decide whether any activities can begin
at each point in simulated time.

With the activity-scanning approach, a modeler concentrates on the activities of a

model and those conditions, simple or complex that allows an activity to begin. At
each clock advance, the conditions for each activity begin. Proponents claim that the
activity-scanning approach is simple in concept and leads to modular models that are
more maintainable and easily understood and modified by other analysts at later
times. They admit, however, that the repeated scanning to determine whether an
activity-scanning approach has been modified (and made conceptually somewhat
more complex) by what is called the three-phase approach, which combines some of
the features of event scheduling with activity scanning to allow for variable time
advance and the avoidance of scanning when it is not necessary, but keeping the
main advantages of the activity-scanning approach.

In the three-phase approach, events are considered to be activities of duration-

zero time units. With this definition, activities are divided into two categories, called B
and C.

B activities Activities bound to occur; all primary events and unconditional activities.

C activities Activities or events that are conditional upon certain conditions being
true.

The B-type activities and events can be scheduled ahead of time, just as in the
event-scheduling approach. This allows variable time advance. The FEL contains only B-
type events. Scanning to check if any C-type activities can begin or C-type events occur
happens only at the end of each time advance, after all B-type events have completed. In
summary, with the three-phase approach, the simulation proceeds with repeated
execution of the three phases until it is completed:

Phase A Remove the imminent event from the FEL and advance the clock to its
event time. Remove any other events from the FEL that have the same event time.

Phase B Execute all B-type events that were removed from the FEL. (This may free
a number of resources or otherwise change system state.)
 27

Phase C Scan the conditions that trigger each C-type activity and activate any whose
conditions are met. Rescan until no additional C-type activities can begin or events
occur.
The three-phase approach improves the execution efficiency of the activity
scanning method. In addition, proponents claim that the activity-scanning and three-
phase approaches are particularly good at handling complex resource problems, in which
various combinations of resources are needed to accomplish different tasks. These
approaches guarantee that resources being freed at a given simulated time will be freed
before any available resources are reallocated to new tasks.

List processing

List processing deals with methods for handling

lists of entities & the failure event list. Simulation
packages provide both explicitly for an analysts use as
well as hidden in the simulation machanism behind the
language, facilities for an analyst or the model itself to
use lists & perform the basic operations on lists.
Section 3.2.1 describes the basic properties and operations performed on lists.
Section 3.2.2 discusses the use of arrays for processing lists and the use or array indices
to create linked lists, arrays being a simpler mechanism for describing the basic
operations than the more general dynamically allocated linked lists discussed in Section
3.2.3. Finally, Section 3.2.4 briefly introduces some of the more advanced techniques for
managing lists.

The purpose of this discussion of list processing is not to prepare the reader to
implement lists and their processing in a general-purpose language such as FORTRAN,
C, or C++ but rather to increase the reader’s understanding of lists and the underlying
concepts and operations.

3.2.1. Lists: Basic Properties and Operations

As previously discussed, lists are a set of ordered or ranked records. In

simulation, each record represents one entity or one event notice.

Since lists are ranked, they have a top or head (the first item on the list); some
way to traverse the list (to find the second, third, etc, items on the list); and a bottom or
 28

tail (the last item on the list). A head pointer is a variable that points to or indicates the
record at the top of the list. Some implementations of lists may also have a tail pointer
that points to the bottom item on the list.

For purposes of discussion, an entity along with its attributes or an event notice
will be referred to as a record. An entity identifier and its attributes are fields in the entity
record; the event type, event time, and any other event-related data are fields in the
event-notice record. Each record on a list will also have a field that holds a “next pointer”
that points to the next record on the list, providing a way to traverse the list. Some lists
may also require a “previous pointer” to allow traversing the list from bottom to top.
For either type of list, the main activities in list processing are adding a record to
a list and removing a record from a list. More specifically, the main operations on a list
are:

1. Removing a record from the top of the list.

2. Removing a record from any location on the list.
3. Adding an entity record to the top or bottom of the list.
4. Adding a record to an arbitrary position on the list, determined by the ranking
rule.

While the first and third operations, removing or adding a record to the top or
bottom of the list, can be carried out in minimal time by adjusting two record pointers and
the head or tail pointer, the other two operations require at least a partial search through
the list. Making these two operations efficient is the goal of list- processing techniques.

In the event-scheduling approach, when time is advanced and the imminent

event is due to be executed, the removal operation takes place first, namely, the event at
the top of the FEL is removed from the FEL. If an arbitrary event is being canceled, or an
entity is removed from a list based on some of its attributes (say, for example, its priority
and due date) to begin an activity, then the second removal operation is performed.
When an entity joins the back of a first-in, first-out queue implemented as a list, then the
third operation, adding an entity to the bottom of a list, is performed. Finally, if a queue
has a ranking rule of earliest due date first, then upon arrival to the queue, an entity must
be added to the list, not at the top or bottom, but at the position determined by the due-
date ranking rule.

When simulating on a computer, whether using a general-purpose language

such as FORTRAN, C, or C++, or a simulation package, each entity record and event
notice is stored in a physical location in computer memory. There are two basic
possibilities: (a) All records are stored in arrays. Arraya hold successive records in
contiguous locations in computer memory. They therefore can be referenced by an array
index that can be thought of as a row number in a matrix. (b) All entities and event
notices are represented by structures (as in C) or classes (as in C++), allocated from
RAM as needed, and tracked by pointers to a record or structure.
 29

Most simulation packages use dynamically allocated records and pointers to

keep track of lists of items. As arrays are conceptually simpler, the concept of linked lists
is first explained through arrays and array indices in Section 3.2.2 and then applied to
dynamically allocated records and pointers in Section 3.2.3.

3.2.2. Using Arrays For List Processing

The array method of list storage is typical of FORTRAN but may be used in other
procedural languages. As most versions of FORTRAN do not have actual record-type
data structures, a record may be implemented as a row in a two-dimensional array or as
a number of parallel arrays. For convenience, we use the notation R(i) to refer to the i th
record in the array, however it may be stored in the language being used. Most modern
simulation packages do not use arrays for list storage but rather use dynamically
allocated records—that is, records that are created upon first being needed and
destroyed when they are no longer needed.

Arrays are advantageous in that any specified record, say the i th, can be
retrieved quickly without searching, merely by referencing R(i). Arrays are disadvantaged
when items are added to the middle of a list or the list must be rearranged. In addition,
arrays typically have a fixed size, determined at compile time or upon initial allocation
when a program first begins to execute. In simulation, the maximum number of record for
any list may be difficult or impossible to determine ahead of time, while the current
numberin a list may vary widely over the course of the simulation run. Worse yet, most
simulations repuire more than one list, and if kept in separate arrays, each would have to
be dimensioned to the largest the list would ever be, potentially using excessive amounts
of computer memory.

3.2.3. Using Dynamic Allocation and Linked Lists

In procedural languages such as C and C++, and in most simulation languages,

entity records are dynamically created when an entity is created and event notice records
are dynamically created whenever an event is scheduled on the future event list. The
languages themselves, or the operating systems on which they are running, maintain a
linked list of free chunks of computer memory and allocate a chunk of desired size upon
request to running programs. (Another use of linked lists!) When an entity “dies” – that is,
exits from the simulated system – and also after an event occurs and the event notice is
no longer needed, the corresponding records are freed, making that chunk of computer
memory available for later reuse; the language or operating system adds the chunk to the
list of free memory.

In this text, we are not concerned with the details of allocating and freeing
computer memory, and we will assume that the necessary operations occur as needed.
With dynamic allocation, a record is referenced by a pointer instead of an array index.
When a record is allocated in C or C++, the allocation routine returns a pointer to the
allocated record, which must be stored in a ariable or a field of another record for later
 30

use. A pointer to a record can be thought of as the physical or logical address in

computer memory of the record.

In our example, we will use a notation for records identical to that in the previous
section (3.2.2):

Entities: [ID, attributes, next pointer]

Event notice: [event type, event time, other data, next pointer]

but we will not reference them by the array notation R(i) as before, because it would be
misleading. If for some reason we wanted the third item on the list, we would have to
traverse the list, counting items until we reached the third record. Unlike arrays, there is
no way to retrieve directly the ith record in a linked list, as the actual records may be
stored at any arbitrary location in computer memory and are not stored contiguously as
are arrays.

Summary

This chapter introduced the major concepts & building blocks in simulations the
most important being entities & attributes, events & activities.

Exercises

1) Prepare a banking system model with the future event list if the 4 customers & 2
solvers and model the clock for 2 days
2) Prepare 1. System state 2. System entities & their attributes 3. Sets & the entities
that may be put into the sets. 4. Events & activities for any 5 systems (either single
server or double server).
3) Prepare a table in the until the clock reaches time 15, _________the international &
service times given below in the order shown. The stopping event will be at time 30.
International times 1 5 6 3 8
Service time 5 3 4 1 5.





 31

4
Simulation Software

In this chapter we first discuss the history of simulation software – a history that
is just reaching middle age. We base this history on our collective experience, articles
written by Richard Nance, and panel discussions at the annual Winter Simulation
Conference.

Next, we discuss features and attributes of simulation software. If you were about
to purchase simulation software, what would concern you? Would it be the cost, the ease
of learning, the ease of use, or would it be the power to model the kinds of system with
which you are concerned? Or would it be the animation capabilities? We then discuss
other issues and concerns related to the selection of simulation software.

Software that is used to develop simulation models can be divided into three
categories. First, there are the general – purpose programming languages, such as
FORTRAN, C, and C++. Second, there are simulation programming languages,
TM
examples being GPSS/H , and SIMAN V. Third, there are the simulation environments.
This category includes many products that are distinguished one way or another (by, for
example, cost, application area, or type of animation) but have common characteristics
such as a graphical user interface and an environment that supports all (or most) aspects
of a simulation study. Many simulation environments contain a simulation programming
language, while some take a graphical approach similar to process flow diagramming.

In the first category, we discuss simulation in C++. C++ is a general purpose

programming language that was not specifically designed for use in simulation. C++ was
chosen since it is widely used and widely available. Today very few people are writing
discrete-event simulation models using programming languages alone. However, in
certain application areas, some people are

4.2 Selection of Simulation Software

This chapter includes a brief introduction to a number of simulation software packages.

There are guides to simulation software published in at least two sources. In May of each
year, IIE Solutions (Institute of Industrial Engineers) presents a “Simulation Software
Buyer’s Guide.” The 1999 guide had 39 products in it. Every two years, OR/MS Today
publishes a simulation software survey [Swain, 1999]. The 1999 issue had 54 products,
including simulation support packages such as input data analyzers.
 32

There are many features that are relevant when selecting simulation software [Banks,
1996]. Some of these features are shown, along with a brief description, in Tables 4.1 to
4.5. We offer the following advice when evaluating and selecting simulation software:

1. Do not focus on a single issue such as ease of use. Consider the accuracy and
level of detail obtainable, ease of learning, vendor support, and applicability to
your problems.

2. Execution speed is important. Do not think exclusively in terms of experimental

runs that take place at night and over the weekend. Speed affects development
time. During debugging, an analyst may have to wait for the model to run up to
the point in simulated time where an error occurs many times before the error is
identified.

3. Beware of advertising claims and demonstrations. Many advertisements exploit

positive features of the software only. Similarly, the demonstrations solve the test
problem very well, but perhaps not your problem.

4. Ask the vendor to solve a small version of your problem.

5. Beware of “checklists” with “yes” and “no” as the entries. For example, many
packages claim to have a conveyor entity. However, implementations have
considerable variation and level f fidelity. Implementation and capability are what
is important. As a second example, most packages offer a runtime license, but
these vary considerably in price and features.

6. Simulation users ask if the simulation model can link to and use code or routines
written in external languages such as C, C++, or FORTRAN. This is a good
feature, especially when the external routines already exist ad are suitable for the
purpose at hand. However, the more important question is whether the simulation
package and language are sufficiently powerful to avoid having to write logic in
any external language.

7. There may be a significant trade-off between the graphical model-building

environments and ones based on a simulation language. While graphical model
building removes the learning curve due to language syntax, it does not remove
the need for procedural logic in most real-world models and the debugging to get
it right. Beware of “no programming required” unless either the package is a near-
perfect fit to your problem domain, or programming (customized procedural
logic) is possible with the supplied blocks, nodes, or process flow diagram, in
which case “no programming required” refers to syntax only and not the
development of procedural logic.
 33

4.3 An Example Simulation

EXAMPLE 4.1 (The Checkout Counter: A Typical Single-Server Queue)

The system, a grocery checkout counter, is modeled as a single – server queue. The
simulation will run until 1000 customers have been served. In addition, assume that the
inter arrival times of customers are exponentially distributed with a mean of 4.5 minutes,
and that the service times are (approximately) normally distributed with a mean of 3.2
minutes and a standard deviation of 0.6 minute. (The approximation is that service times
are always positive). When the cashier is busy, a queue forms with no customers turned
away. This example was manually simulated in Examples 3.3 and 3.4 using the event
scheduling point of view. The model contains two events, the arrival and departure
events. Figures 3.5 and 3.6 provide the event logic.

The following three sections illustrate the simulation of this single – server queue in C ++,
GPSS/H, and CSIM. Although this example is much simpler than models that arise in the
study of complex systems, its simulation contains the essential components of all
discrete-event simulations.

4.4 Simulation in C ++

C ++ is a widely used programming language that has been used extensively in

simulation. (The definitive book on C ++ is its designer’s [Stroustup, 1997]. It does not,
however, provide any facilities directly aimed at aiding the simulation analyst, who
therefore must program all details of the event – scheduling/time – advance algorithm,
the statistics – gathering capability, the generation of samples from specified probability
distributions, and the report generator. However, the runtime library does provide a
random – number generator. Unlike FORTRAN or C, the object – orientedness of C ++
does support modular construction of large models. For the most part, the special –
purpose simulation languages hide the details of event scheduling, whereas in C ++ all
the details must be explicitly programmed. However, to a certain extent simulation
libraries such as CSIM [Schwetman, 1987] alleviate the development burden by providing
access to standardized simulation functionality and hiding low-level scheduling minutiae,

Any discrete-event simulation model written in C ++ contains the components discussed

in Section 3.1: system state, entites and attributes, sets, events, activities and delays,
plus the component listed below. To facilitate development and debugging, it is best to
organize the C ++ model in a modular fashion using subroutines. The following
components are common to almost all models written in C ++:

Clock A variable defining simulated time

Initialization subroutine A routine to define the system state at time 0
Min-time event subroutine A routine that identifies the imminent event – that is, the
elements of the future event list (Future Event List) which has the smallest time – stamp
 34

The simulation begins by setting the simulation Clock to zero, initializing cumulative
statistics to zero, generating any initial events (there will always be at least one) and
placing them on the Future Event List, and defining the system state at time 0. The
simulation program then cycles, repeatedly passing the current least-time event to the
appropriate event subroutines until the simulation is over. At each step, after finding the
imminent event but before calling the event subroutine, the simulation Clock is advanced
to the time of the imminent event. (Recall that during the simulated time between the
occurrence of two successive events, the systems state and entity attributes do not
change in value. Indeed, this is the definition of discrete – event simulation: The system
state changes only when an event occurs.) Next, the appropriate event subroutine is
called to execute the imminent event, update cumulative statistics, and generate future
events (to be placed on the Future Event List). Executing the imminent event means that
the system state, entity attributes, and set membership are changed to reflect the fact
that the event has occurred. Notice that all actions in an event subroutine takes places at
one instant of simulated time. The value of the variable Clock does not change in an
event routine. If the simulation is not over, control passes again to the time-advance
subroutine, then to the appropriate event subroutine, and so on. When the simulation is
over, control passes to the report generator, which computes the desired summary
statistics from the collected cumulative statistics and prints a report.

The efficiency of a simulation model in terms of computer runtime is determined to a

large extent by the techniques used to manipulate the Future Event List and other sets.
As discussed in Section 3.1, removal of the imminent event and addition of a new event
are the two main operations performed on the Future Event List. Virtually every C ++
system now includes the “C ++ standard libraries,” whose priority-queue data structure
we will use in this example to implement the event list and whose generic queue we will
use to implement the list of waiting customers. The underlying priority-queue organization
is efficient, in the sense of having access costs that grow only in the logarithm of the
number of elements in the list.

EXAMPLE 4.2 (Single-Server Queue Simulation C ++)

The grocery checkout counter, defined in detail in Example 4.1, is now simulated using C
++. A version of this example was simulated manually in Examples 3.3 and 3.4, where
the systems state, entities and attributes, sets, events, activities, and delays were
analyzed and defined.

Class Event represents an event. It stores a code for the event type (arrival or departure),
and the event time – stamp. It has associated methods (functions) for creating an event
and accessing its data. It also has associated special functions, called operators, which
provide semantic meaning to relational operators < and = = between events. These are
used by the C ++ standard library queue implementations. The subroutines for this model
and the flow of control are shown in Figure 4.2, which is an adaptation of Figure 4.1
 35

be used to model any situation where transactions (entities, customers, units of traffic)
are flowing through a system (e. g., a network of queues, with the queues preceding
scarce resources). The block diagram is converted to block statements, control
statements are added, and the result is a GPSS model.

The first version of GPSS was released by IBM about 1961. Since is was the first process
– interaction simulation language, and due to its popularity, it has been implemented
anew and improved by many parties since 1961, with GPSS/H being the most widely
used version in use today. Example 4.3 is based on GPSS/H.

GPSS/H is a product of Wolverine Software Corporation, Annandale, VA (Banks, Carson,

and Sy, 1995; Crain and Henriksen, 1999). It is a flexible, yet powerful tool for simulation.
Unlike the original IBM implementation, GPSS/H includes built-in file and screen I/o, use
of an arithmetic expression as a block operand, an interactive debugger, faster execution,
expanded control statements, ordinary variables and arrays, a floating-point clock, built-in
math functions, and built – in random – variate generators.
TM
The animator for GPSS/H is Proof Animation , another product of Wolverine Software
Corporation (Henriksen, 1999). Proof Animation provides a 2-D animation, usually based
on a scale drawing. It can run in post processed mode (after the simulation has finished
running) or concurrently. In post processed mode, the animation is driven by two files: the
layout file for the static background, and a trace file that contains commands to make
objects move and produce other dynamic events. It can work with any simulation
package that can write the ASCII trace file. Alternately, it can run concurrently with the
simulation by sending the trace-file commands as messages, or can be controlled directly
when using its DLL (dynamic link library) version.

EXAMPLE 4.3 (Single-Server Queue Simulation in GPSS/H)

Figure 4.10 exhibits the block diagram and figure 4.11 the GPSS program for the grocery
store checkout –counter model described in Example 4.2. Note that the program (Figure
4.11) is a translation of the block diagram together with additional definition and control
statements.

In figure 4.10, the GENERATE block represents the arrival event, with the interarrival
times specified by RVEXPO (1, &IAT). RVEXPO stands for “random variable,
exponentially distributed,” the 1 indicates the random-number stream to use, and &IAT
indicates that the mean time for the exponential distribution comes from a so-called
ampervariable &IAT. Ampervariable names begin with the “&” character; Wolverine
added ampervariables to GPSS because the original IBM implementation had limited
support for ordinary global variables, with no user freedom for naming them. (In the
discussion that follows all nonreserved words are shown in italics.)

The next block is a QUEUE with a queue named SYSTIME. It should be noted that the
QUEUE block is not needed for queues or waiting lines to form in GPSS. The true
 36

purpose of the QUEUE block is to work in conjunction with the DEPART block to collect
data on queues or any other subsystem.

Figure 4.10. GPSS block diagram for single – server queue simulation.
 37

In Example 4.3, we want to measure the system response time – that is, the time a
transaction spends in the system. By placing a QUEUE block at the point where
transactions enter the system, and the counter part of the QUEUE block, the DEPART
block, at the point where the transactions complete their processing, the response times
will be automatically collected. The purpose of the DEPART block is to signal the end of
data collection for an individual

SIMULATE

Define Ampervariables

INTEGER &LIMIT
REAL &IAT, &MEAN,&STDEV, &COUNT
LET &IAT = 4.5
LET &MEAN = 3.2
LET &STDEV = .6
LET &LIMIT = 1000

Write Input Data to File

PUTPIC FILE = OUT, LINES = 5, (&IAT, &MEAN, &STDEV, &LIMIT)

Mean interarrival time **.** minutes

Mean service time **.** minutes
Standard deviation of service time **.** minutes
Number of customers to be served *****

GPSS/H Block Section

GENERATE RVEXPO (1, &IAT) Exponential arrivals

QUEUE SYSTIME Begin response time data collection
QUEUE LINE Customer joins waiting line
SEIZE CHECKOUT Begin checkout at cash register
DEPART LINE Customer starting service leaves queue
ADVANCE RVNORM(1,&MEAN, Customer’s service time
&STDEV)
RELEASE CHECKOUT Customer leaves checkout area
DEPART SYSTIME End response time data collection
TEST GE M1, 4, TER Is response time GE 4 minutes?
BLET &COUNT =&COUNT+1 If so, add 1 to counter

TER TERMINATE 1
 38

START &LIMIT Simulate for required number

Write Customized Output Data to File

PUTPIC FILE = OUT, LINES = 7, (FR (CHECKOUT) / 1000, QM (LINE), -

QT (SYSTIME), &COUNT/N(TER), ACL, N(TER))

Server utilization .***

Maximum line length **
Average response time **.** minutes
Proportion who spend four minutes .***
Or more in the system
Simulation run length ****.** minutes
Number of departures ****

END

Figure 4.11 GPSS/H program for single – server queue simulation.

transaction. The QUEUE and DEPART block combination is not necessary for queues to
be modeled, but rather is used for statistical data collection.

The next QUEUE block (with name Line) begins data collection for the waiting –
line before the cashier. The customers may or may not have to wait for the cashier. Upon
arrival to an idle checkout counter or after advancing to the head of the waiting line, a
customer captures the cashier as represented by the SEIZE block with the resource
named CHECKOUT. Once the transaction representing a customer captures the cashier
represented by the resource CHECKOUT, the data collection for the waiting line statistics
ends, as represented by the DEPART block for the queue named LINE. The transaction’s
service time at the cashier is represented by an ADVANCE block. RVNORM indicates
“random variable, normally distributed.” Again, random – number stream 1 is being used,
the mean time for the normal distribution is given by ampervariable & MEAN, and its
standard deviation is given by ampervariable & STDEV. Next,/ the customer gives up the
use of the facility CHECKOUT with a RELEASE block. The end of the data collection for
response times is indicated by the DEPART block for the queue SYSTIME.

Next, there is a TEST block that checks to see if the time in the system, M1, is
greater than or equal to 4 minutes. (Note that M1 is a reserved word in GPSS/H; it
automatically tracks transaction total time in system.) In GPSS/H, the maxim is “if true,
pass through.” Thus, if the customer has been in the system 4 minutes or longer, the next
BLET block (for block LET) adds one to he counter & COUNT. If not true, the escape
route is to the block labeled TER. That label appears before the TERMINATE block,
whose purpose is the removal of the transaction from the system. The TERMINATE block
 39

has a value “1” indicating that one more transaction is added toward the limiting value, or
“transactions to go.”

The control statements in this example are all of those lines in Figure 4.11 that
precede or follow the block section. (There are eleven blocks in the model from the
GENERATE block to the TERMINATE block). The control statements that begin with an
asterisk are comments, some of which are used for spacing purposes. The control
statement SIMULATE tells GPSS/H to conduct a simulation; if omitted, GPSS/H compiles
the model and checks for errors only. The ampervariables are defined as integer or real
by control statements INTEGER and REAL. It seems that the ampervariable & COUNT
should be defined as an integer; however, it will be divided later by the number of
customers to obtain a proportion. If it were integer, the result of an integer divided by an
integer would be truncation, and that is not desired in this case. The four assignment
statements (LET) provide data for the simulation. These four values could have been
placed directly in the program; however, the preferred practice is to place them in
ampervariables at the top of the program, so that changes can be made more easily, or
the model modified to read them from a data file.

To insure that the model data is correct, and for the purpose of managing
different scenarios simulated. It is good practice to echo the input data. This is
accomplished with a PUTPIC (for “put picture”) control statement. The five lines following
PUTPIC provide formatting information, with the asterisks being markers (called picture
formatting) in which the values of the four ampervariables replace the asterisks when
PUTPIC is executed. Thus, “**.**” indicates a value that may have two digits following the
decimal point.

Mean interarrival time 4.50 minutes

Mean service time 3.20 minutes
Standard deviation of service time 0.60 minutes
Number of customers to be served 1000
Server utilization 0.676
Maximum line length 7
Average response time 6.33 minutes
Proportion who spend four minutes 0.646
Or more in the system
Simulation runlength 4767.27 minutes
Number of departures 1000

Figure 4.12 Customized GPSS/H output for single-server queue simulation.

The START control statement controls simulation execution. It starts the

simulation, sets up a “termination to-go” counter with initial value its operand (&LIMIT),
and controls the length of the simulation.
 40

After the simulation completes, a second PUTPIC control statement is used to

write the desired output data to the same file OUT. The printed statistics are all
automatically gathered by GPSS. The first output in the parenthesized list is the server
utilization. FR (CHECKOUT)/1000 indicates that the fractional utilization of the facility
CHECKOUT is printed. Since FR (CHECKOUT) is in parts per thousand, the
denomination is provided to computer fractional utilization. QM(LINE) is the maximum
value in the queue LINE during the simulation. QT(SYSTIME) is the average time in the
queue SYSTIME. & COUNT/N(TER) is the number of customers who had a response
time of four or more minutes divided by the numbers of customers that went through the
block with label TER, r N (TER). AC1 is the clock time, whose last value gives the length
of the simulation.

The contents of the custom output file OUT are shown in Figure 4.12. The
standard GPSS/H output file is displayed in Figure 4.13. Although much of the same data
shown inj the file OUT can be found in the standard GPSS/H output, the custom file is
more compact and uses the language of the problem rather than GPSS jargon. There are
many other reasons that customized output files are useful. For example, if 50
replications of the model are to be made and the lowest, highest, and average value of a
response are desired, this can be accomplished using control statements with the results
in a very compact form, rather than extracting the desired values from 50 standard output
reports.

4.6 Simulation in CSIM

CSIM is a system for using the C or C ++ language for modeling, along with a
rich library of predefined objects to support process-oriented simulation modeling. It is an
exepensive commercial product sold and maintained by Mesquite. Software [Shwetman,
1996]; it has a wide user base in industry and academia, principally to model computer
and communication systems. CSIM is fast, owing to careful implementation and being a
compiled language.

* quired customers, it calls the wait function associated with CSIM event done. This
causes it to suspend until done is set by the last customer process of interest. The call to
done. Set ( ) in turn releases sim to call Report Generation to give its report. Our CSIM
implementation of Report Generation can be essentially that illustrated in Figure 4.7, with
the exception of calling CSIM provided facility routine util to get server utilization, and
resp to get the average response time :

double RHO = f. util ( );

double AVGR = f. resp ( )

CSIM bridges the gap between models developed in pure C ++ and models
developed in languages specifically designed for simulation. It provides the flexibility
offered by a general programming language, with essential support for simulation.
 41

4.7 Simulation Packages

All the simulation packages described in later subsections run on a PC under

Microsoft Windows 95/98 and/or NT. Some also run on various UNIX workstations and a
few on Apple machines. Although in terms of specifics the packages all differ, generally
they have many things in common.

Common characteristics include a graphical user interface, animation, and

automatically collected outputs to measure system performance. In virtually all packages,
simulated results may be displayed in tabular or graphical form in standard reports and
interactively while running a simulation. Outputs from different scenarios can be
compared graphically or in tabular form. Most provide statistical analyses that include
confidence intervals for performance measures and comparisons, plus a variety of other
analysis methods. Some of the statistical analysis modules are described in Section 4.8.

All the packages described here take the process – interaction world view. A few
also allow event scheduling and mixed discrete – continuous models. For animation,
some emphasize scale drawings in 2 – D or 3-D, while other emphasize iconic – type
animations based on schematic drawings or process flow diagrams. A few offer both
scale-drawing and schematic – type animations. Almost all offer dynamic business
graphing in the form of timelines, bar charts, and pie charts.

4.7.1 Arena

Arena Business, Standard, and Professional Editions are offered by Systems

Modeling Corporation [Sadowski and Bapat, 1999]. Arena can be used for simulating
discrete and continuous systems. A recent addition to the Arena family of products is
OptQeust for Arena, an optimization software package (discussed in Section 4.8.2).

The Arena Business Edition is targeted at modeling business processes and

other systems in support of high-level analysis needs. It represents process dynamics in
a hierarchical flow chart and stores system information in data spreadsheets. It has built
– in activity-based costing and is closely integrated with the flow-charting software Visio .
R

The Arena Standard Edition is designed for more detailed models of discrete and
continuous systems. First released in 1993, Arena employs an object based design for
entirely graphical model development. Simulation models are built using graphical objects
called modules to define system logic and physical components such as machines,
operators, and clerks. Modules are represented by icons plus associated data entered in
a dialog window. These icons are connected to represent entity flow. Modules are
organized into collections called templates. The Arena template is the core collection of
modules providing general – purpose features for modeling all types of applications. In
 42

addition, to standard features, such as resources, queues, process logic, and system
data, the Arena template includes modules focused on specific aspects of manufacturing
and material handling systems. Arena SE can also be used to model combined discrete/
continuous systems, such as pharmaceutical and chemical production, through its built-in
continuous modeling capabilities.

The Arena Professional Edition enhances Arena SE with the capability to craft
custom simulation objects that mirror components of the real system, including
terminology, process logic, data, performance metrics, and animation. The Arena family
also includes products designed specifically to model call centers and high-speed
production lines, namely Arena Call Center and Arena Packaging.

At the heart of Arena is the SIMAN simulation language. For animating simulation
models, Arena’s core modeling constructs are accompanied by standard graphics for
showing queues, resource status, and entity flow. Arena’s 2-D animations are created
using Arena’s built-in drawing tools and by incorporating clipart, AutoCAD, Visio, and
other graphics.

Arena’s Input Analyzer automates the process of selecting the proper distribution
and its parameters for representing existing data, such as process and interarrival times.
The Output Analyzer and Process Analyzer (discussed in Section 4.8.2) automate
comparison of different design alternatives.

4.7.2 Auto Mod

The Auto Mod Product Suite is offered by Auto Simulations [Rohrer, 1999]. It
includes the Auto Mod simulation package, AutoStat for experimentation and analysis,
and AutoView for making AVI movies of the built-in-3D animation. The main focus of the
AutoMod simulation product is manufacturing and material handling systems.

Auto Mod has built – in templates for most common material handling systems,
including vehicle systems, conveyors, automated storage and retrieval systems, bridge
cranes, power and free conveyors, and kinematics for robotics. With its Tanks and Pipes
module, it also supports continuous modeling of fluid and bulk material flow.

The Path Mover vehicle system can be used to model lift trucks, humans walking
or pushing carts, automated guided vehicles, trucks and cars. All the movement
templates are based on a 3-D scale drawing (drawn or imported from CAD as 2-D or 3-
D). All the components of a template are highly paramterized. For example, the conveyor
template contains conveyor sections, stations for load induction or removal, motors, and
photo-eyes. Sections are defined by length, width, speed, acceleration and type
(accumulating or nonaccumulating), plus other specialized parameters. Photo-eyes have
blocked and cleared timeouts that facilitate modeling of detailed conveyor control logic.
 43

In addition to the material handling templates, Auto Mod contains a full simulation
programming language. Its 3-D animation can be viewed from any angle or perspective in
real time. The user can freely zoom, pan, or rotate the 3-D world.

An Auto Mod model consists of one or more systems. A system can be either a
process system, in which flow and control logic are defined, or a movement system
based on one of the material handling templates. Each model must contain one process
system and may contain any number of movement systems. Processes can contain
complex logic to control the flow of either manufacturing materials or control messages,
to contend for resources, or to wait for user – specified times. Loads can move between
processes with or without using movement systems.

In the Auto Mod world view, loads (products, parts, etc.) move from process to
process and compete for resources (equipment, operators, vehicles and queues). The
load is the active entity, executive action statements in each process. To move between
processes, loads may use a conveyor or vehicle in a movement system.

Auto Stat, described in Section 4.8.2, works with Auto Mod models to provide a complete
environment for the user to define scenarios, conduct experimentation, and perform
analyses. It offers optimization based on an evolutionary strategies algorithm.

4.7.3 Deneb/QUEST

QUEST is offered by Deneb Robotics [Donald, 1998]. Deneb also offers a

number of workcell simulators including IGRIP for robotics simulation and programming
and ERGO Sim for ergonomics analyses. QUEST (Queuing Event Simulation Tool) is a
manufacturing-oriented simulation package. Robots and workcells simulated in IGRIP
and ERGO Sim can be imported into QUEST models.

QUEST models are based on 3-D CAD geometry. QUEST combines a simulation
environment with graphical user interface and material flow modules for labor, conveyors,
automated guided vehicles, kinematics, power and free conveyors, and automated
storage and retrieval systems.

For Deneb/IGRIP users, existing robot work cell models and robot or other
complex machinery cycle times can be directly imported to Deneb/QUEST. Similarly,
model of human operators in a workcell developed in Deneb\ERGO Sim can be
incorporate into a QUEST process model both visually and numerically.

A QUEST model consists of elements from a number of element classes. Built-in

element classes include AGV and transporters, buffer, conveyor, labor, machine, parts
and process. Each element has associated geometric data and parameters that define its
behavior. Parts may have a route and control rules to govern part flow. Commonly
needed behavior logic is selected from comprehensive menus, many parameter-driven.
For unique problems, Deneb/QUEST’s Simulation Control Language can be used. This
 44

structured simulation programming language provides distributed processing with access

to all system variables. SCL allows expert users to define custom behaviors and to gain
control over the simulation.

Deneb/QUEST’s open architecture allows the advanced user to perform batch

simulation runs to automatically collect and tabulate data using the Batch Control
Language (BCL). Replications and parameter optimization are controlled with batch
common files.

4.7.4 Extend

Extend offered by Imagin That, Inc. [Krahl, 1999]. Extend combines a block
diagram approach to model building along with an authoring environment for creating
new blocks or complete vertical market simulators. It is process oriented but it also
capable of continuous and combined modeling. It provides iconic animation of the block
diagram plus an interface to Proof Animation [Henriksen, 1999] from Wolverine Software
for 2-D scale – drawing animation.

Elemental blocks in Extend include Generator (for arrivals), Queue, Activity,

Resource Pool, and Exit. In Extend, the active entities, called items, are created at
Generator blocks and move from block to block by way of connectors. Connectors are
also used for “value flows” – that is, for attaching a calculation to a block parameter or
retrieving statistical information from a block for reporting purpose.

Each block has an icon and encapsulates code, parameters, user interface, and
online help. Extend comes with a large set of elemental blocks; in addition, libraries of
blocks for general and specific application areas, such as manufacturing and BPR
(business process reengineering), are available. Third-party developers have created
Extend libraries for vertical market applications, including supply chain analysis chemical
processing, pulp and paper processing, and radio-frequency analysis.

End users can build models by placing and connecting blocks and filling in the
parameters on the dialog window associated with a block. Collections of blocks can be
grouped into a hierarchical block representing a submodel such as a particular machine,
a workstation, or a subassembly line. In this way, a higher-level block diagram can be
displayed to the end user, hiding many of the details. Parameters from the submodel
network of blocks that an end user desires to control can be grouped and displayed at the
level of the hierarchical block. Using this capability, a firm could build a set of customized
blocks representing various components of their manufacturing facility.

For creating new blocks, Extend comes with a compiled C-like simulation
programming language called Modl. It contains simulation support, as well as support for
custom user interfaces and message communication. Extend has an open architecture,
so that in most cases the source code for blocks is available to advanced model builders.
 45

The statistics library of blocks is used to collect and analyze output data, such as
computing confidence intervals.

4.7.5 Micro Saint

Micro Saint is offered by Micro Analysis and Design, Inc. [Bloechle and
Laughery, 1999]. With Micro Saint, a model builder develops a model by creating a flow-
chat diagram to describe a network of tasks. The animation is iconic and is based on the
flow-chart representation. Through menus and parametrized icons, the network can have
branching logic, sorted queues, and conditional task execution. Micro Saint is used in
manufacturing, health care, retail, and military applications. It supports ergonomics and
human performance modeling. It works with Opt Quest to provide optimization capability.

4.7.6 Pro Model

Pro Model is offered by PROMODEL Corporation [Price and Harrell, 1999]. It is a

simulation and animation tool designed to model manufacturing systems. The company
also offers MedModel for health care systems, Service Model for service systems, and
Process Model, a flow – chart – based simulation product for business processes. Pro
Model offers 2-D animation with an optional 3-D- like perspective view.

Pro Model has manufacturing – oriented modeling elements and rule – based
decision logic. Some systems can be modeled by selecting from Pro Model’s set of highly
parameterized modeling elements. In addition, its simulation programming language
provides for modeling special situations not covered by the built- in choices.

The modeling elements in Pro-Model are parts or entities, locations, resources,

path networks, routing and processing logic, and arrivals. Parts arrive and follow the
routing and processing logic from location to location. Resources are used to represent
people, tools or vehicles that transport parts between locations, perform an operation on
a part at a location, or perform maintenance on a location or other resource that is down.
Resources may travel on path networks with given speeds, accelerations, and pick up
and setdown times. The routing and processing element allows user – defined procedural
logic in Pro – Model’s simulation programming language.

Pro Model’s runtime interface allows a user to define multiple scenarios for
experimentation. Sim Runner (discussed in Section 4.8.2.) adds the capability to perform
an optimization; it is based on an evolutionary strategy algorithm, a variant of the genetic
algorithm approach.

4.7.7 Taylor ED

Taylor Enterprise Dynamics (ED) is offered by F & H Simulations [Hullinger,

1991]. It is built on different concepts than its predecessor package, Taylor II. Taylor ED
 46

is used to model manufacturing, werehousing, and material handling processes plus

service and data flow processes. It can also be used to monitor flow processes in real
time. It offers both 2-D and 3-D animation.

Taylor ED is based on the concept of an atom. Atoms are Taylor ED’s smart
objects and model-building resources. Model builders can access standard libraries of
atoms as well as create new atoms using Taylor ED’s Atom Editor. Atoms are used to
represent the products or entities flowing through the system as well as the resources
acting on these entities. In fact, everything in Taylor ED is an atom, whether it is a
resource, a product, a model (or submodel), a table, a report, a graph, a record, or a
library.

To build a model, a user first creates a model layout by selecting atoms from the
atom library and placing them on the screen. Next the user connects the atoms and
defines the routing of products or entities (also atoms) flowing through the model. Next
the use assigns logic, including routing logic, to each atom by editing its parameter fields.

An atom is an object with four dimensions (x, y, z, and time). Each atom can
have a location, speed, and rotation and dynamic behavior over time. Atoms can inherit
their behavior from other atoms and contain other atoms; atoms can be created and
destroyed. Atoms can be viewed simultaneously in 2-D and 3-D animation.

One of the hallmark characteristics of the atom is its reusability. For example, a
workcell can be created by using the Atom Editor or by combining several existing atoms.
The workcell can be saved as an atom, included in a standard atom library, then reused
in a new model or shared with other Taylor ED modelers.

Taylor ED includes both an experiment module and the Opt Quest Optimizer
(discussed in Section 4.8.2).

4.7.8 WITNESS

WITNESS is offered by The Lanner Group [Mehta and Rawles, 1999]. WITNESS
is strongly machine oriented and contains many elements for discrete part manufacturing.
It also contains elements for continuous processing such as the flow of fluids through
processors, tanks and pipes.

WITNESS models are based on template elements. Elements may be combined

into a designer element module to be reused. The machine element can be single, batch,
production, assembly, multisation or multicycle. The behavior of each element is
described on a tabbed detail from in the WITNESS user interface.

It offers a 2-D layout animation plus a process flow view. The 2-D graphics offers
a walk-through capability controlled by a mouse or a camera attached to an entity.
 47

4.8 Experimentation and Statistical Analysis Tools

4.8.1 Common Features

Virtually all simulation packages offer various degrees of support for statistical
analysis of simulation outputs. In recent years, many packages have added optimization
as one of the analysis tools. To support analysis, most packages provide scenario
definition, run management capability, as well as data export to spreadsheet and other
external applications.

Optimization is used to find a “near – optimal” solution. The user must define an
objective or fitness function, usually a cost or costlike function that incorporates the trade
– off between additional throughput and additional resources. Until recently, the methods
available for optimizing a system had difficulty coping with the random and nonlinear
nature of most simulation outputs. Advances in the field of metaheuristics have offered
new approaches to simulation optimization, based on artificial intelligence, neural
networks, genetic algorithms, evolutionary strategies, tabu search, and scatter search.

4.8.2 Analysis Tools

This section briefly discusses the Arena Output and Process Analyzers AutoStat
for AutoMod, Sim Runner for ProModel, and OptQuest, which is used in a number of
simulation products.

Arena’s Output and Process Analyzer

Arena comes with the Output Analyzer and Process Analyzer. In addition, Arena
uses Opt Quest for optimization.

The Output Analyzer provides confidence intervals, comparison of multiple

systems, and warm up determination to reduce initial-condition biases. It creates various
plots, charts, and histograms, smoothes responses, and performs correlation analysis. To
compute accurate confidence intervals, it does internal batching (both within and across
replication, with no user intervention) and data truncation to provide stationary,
independent and normally distributed data sets.

The Process Analyzer adds sophisticated scenario-management capabilities to

Arena for comprehensive design of experiments. It allows a use to define scenarios,
make the desired runs, and analyze the results. It allows an arbitrary number of controls
 48

and responses. Responses can be added after runs have been completed. It will rank
scenarios by any response and provide summaries and statistical measures of the
responses. A user can view 2-D and 3-D charts of response values across either
replications or scenarios.

AutoStat

AutoStat is the run manager and statistical analysis product in the AutoMod
product family [Rohrer, 1999]. AutoStat provides a number of analyses, including warm
up determination for steady-state analyses, absolute and comparison confidence
intervals, design of experiments, sensitivity analysis, and optimization using an
evolutionary strategy. The evolutionary strategies algorithm used by AutoStat is well
suited to finding a near-optimal solution without getting trapped at a local optimum.

With AutoStat, an end user can define any number of scenarios by defining
factors and their range of values. Factors include single parameters such as resource
capacity or vehicle speed, single cells in a data file, and complete data files. By allowing
a data file to be a factor, a user can experiment with, for example, alternate production
schedules, customer orders for different days, different labor schedules, or any other
numerical inputs typically specified in a data file. Any standard or custom output can be
designated as a response. For each defined response, AutoStat computes descriptive
statistics (average, standard deviation, minimum and maximum) and confidence
intervals. New responses can be defined after runs are made, because AutoStat archives
and compresses the standard and custom outputs from all runs. Various charts and plots
are available to provide graphical comparisons.

AutoStat is capable of distributing simulation runs across a local area network

and pulling back all results to the user’s machine. Support for multiple machines and
CPUs gives users the ability to make many more runs of the simulation tan would
otherwise be possible by using idle machines during off hours. This is especially useful in
multifactor analysis and optimization, both of which may require large numbers of runs.
AutoStat also has a diagnostics capability that automatically detects “unusual” runs,
where the definition of “unusual” is user – definable.

Auto Stat also works with two other products from Auto Simulations: the AutoMod
Simulator, a spreadsheet – based job-shop simulator, and AutoSched AP, a rule – based
simulation package for finite – capacity scheduling in the semi conductor industry.

Opt Quest
 49

Opt Quest was developed by Fred Glover, University of Colorado and co-founder
of Optimization Technologies, Inc. [Glover et al., 1999]. It is available for Arena, CSIM,
Micro Saint, QUEST, and Taylor ED.

Opt Quest is based on a combination of methods: scatter search, tabu search

with adaptive memory, and a neural network component. Scatter search is a population
based approach, in some ways similar to genetic algorithms; existing solutions are
combined to make new solutions. Tabu search is then superimposed to prohibit the
search from reinvestigating previous solutions and neural networks to screen out
solutions likely to be poor.

When using OptQuest, a user defines an objective function to minimize or

maximize. Simulation examples include throughout or a cost function. Opt Quest allows a
user to specify the range of variable parameters as well as constraints on parameters,
responses, and combinations of parameters – in fact, any set of conditions that can be
represented by a mixed-integer programming formulation.

SimRunner

Sim Runner works with ProModel [Prive and Harrell, 1999]. Its optimization is
based on evolutionary strategy, a variant of the genetic algorithm approach. A user
specifies an objective function (a response to minimize or maximize) and the input factors
to vary. In its Stage One Optimization Analysis, Sim Runner performs a factorial design of
experiments to determine which input factors have a significant effect on the objective
function. Those factors that have no significant effect can be eliminated from the search
for an optimum. The Stage Two Simulation Optimization uses the evolutionary strategy
algorithm to conduct a multivariable optimization search.

4.9 Trends in Simulation Software

This section is based on a plenary address given by Banks [2000] to the

European Simulation Multi-Conference 2000. Many of the remarks are reproduced here
with the permission of the Society for Computer Simulation International, European
Council, Ghent, Belgium. Since it is difficult to predict software developments for even
one year, the trends predicted here may be little more than more speculation!

4.9.1 High-Fidelity Simulation

High-fidelity simulation is exceptional graphics quality and very accurate

emulation. The term “Virtual” is often used, implying that the emulation is nearly the same
as that which is being modeled, i.e., it looks and acts like the real thing. However, the
model is in the computer and the actual object is real. In terms of manufacturing and
material handling, consider a virtual machine, a virtual factory, and a virtual warehouse. If
 50

you are looking at a virtual machine on your computer monitor, it looks very much like
that machine. If the machine is updated, it is only necessary to update the simulation
model. You don’t have to spend $1,000,000 for a modern centerless grinder, you just
simulate it in the computer to study its operation. Of course, it you want to make
bearings, you will need to buy the grinder.

Another dimension of high fidelity simulation is to make it selective. You could

have varying levels of detail. If you wanted the model to run fast and the animation and
solution to be approximate, you could set the dial on low. Or, you could set the dial on
high and get the opposite result. You could also set the dial between the two extremes.

4.9.2 Data Exchange Standards

Engineering Animation Incorporated, Inc. (EAI), has developed a simulation interface

called SDX for Simulation Data Exchange. David Sly (1999) of EAI says:

The main function of the tool is to communicate to the simulation

Package all that is known about the layout in an effort to reduce the
Redundant tasks of redefining equipment in the layout importation
To the simulation tool and that equipment’s interconnectivity and
Attributes. This interface adds value in that it can cut considerable
Time from the process of creating detailed layouts in the simulation
tool and defining equipment properties (such as conveyor speeds,
etc.) The interface is best applied to applications that have a
significant amount of automated material handling equipment such
as conveyors, AGVs, fork trucks and cranes, all typically found in
automotive plants and distribution centers.

Several simulation software firms have worked with EAI to accept SDX. This data
exchange format could emerge as a standard. Rather than redraw the conveyor system
with all of its parameters, it is just imported through SDX.

SDX is not the only data exchange format vying for attention. He Worldwide Web
Consortinum (W3C) is developing XML for extensible markup language. XML is a set of
rules, guidelines, or conventions for designing text formats for data such that the files are
easy to generate and read, they are unambiguous, and they are extensible. It aids in
putting information on the Web and in the retrieval of that information.

XML has a large and growing community. A project coordinated through the
Manufacturing Engineering Laboratory at the U. S. National Institute of Standards and
Technology is to “explore the use of XML as a standard for structured document
interchange on the Web, for exchanging complex data objects between tasks in a
distributed workflow application.”
 51

So, groups are developing XML standards for all types of documentation.
Eventually, some group will want to develop XML standards for representation of
manufacturing and material handling systems.

4.9.3 The Internet

The Internet can deliver modeling capability to the user. The interface can be at
the client (local computer), with the models run on a server. However, this server is not in
a local area network (LAN). It is at the software vendor’s or some other host’s site. The
server can always have the latest version of the model. The interface can be in Java,
which is independent of the computer and has applets for displaying results. Currently,
the speed of Java is not sufficient to serve as the simulation engine.

Another way that we can use the Internet is to have models run on many
computers. Thus, if the model is extremely large or if many replications are needed, the
computing power can be greatly extended via the Internet by distributing the
computational needs.

4.9.4 Old Paradigm versus New Paradigm

In the past, models were always constructed with a limited purpose in mind and
to answer specific questions. That would be called the old paradigm. But, it takes time to
build models in response to questions.

The new paradigm is to have models built at different levels of detail, but
without knowing all of the questions to be asked. This will speed up the
response time, since the model already exists. Under the new paradigm, the
models become an asset. The firm with more prebuilt models will be able to
answer questions faster – but may have more out-of-date models. Indeed,
model maintenance becomes a big issue.

4.9.5 Component Libraries

Consider component libraries stored in a neutral file format. This requires that a
format can be designed and accepted by numerous software vendors, which is in itself a
challenge. System providers could be contributors of modules. For example, a robot
manufacturer could contribute a module for a specific type of robot. The result of having
module libraries is more rapid development of models. The modules could be prepared
so that all that they require is parameterization.

4.9.6 Distributed Manufacturing Simulation/High Level

Architecture
 52

Many years ago, the U. S. Department of Defense (DOD) began a development

process to enable various simulations located throughout the world to exchange data
dynamically during simulation of large-scale military exercises. This is known as
distributed simulation. There can be aircraft maneuvers simulated in one computer
model, tanks simulated in another, ground troops in yet another, and the supply chain in
still another. Also, the models may be at different levels of fidelity. The computers hosting
these models are geographically dispersed, but interacting over the Internet.

The technology has matured to the point where it can now be tried in other
domains. This technology is collectively called High Level Architecture (HLA). HLA has
been designed so that the minimum amount of information is exchanged between the
various simulations sufficient to effect clock synchronization and the exchange of data
objects.

4.9.7 Embedded Simulation

Embedded simulation occurs when simulation models are placed in other

software. This already happens in simulation-based scheduling. It is anticipated that there
will be many other cases in which the simulation becomes invisible. For example, in
maintenance management software, a policy could be entered, and a very short time
later, the user would see the anticipate result of that policy. The user might not know that
the output resulted from a simulation.

4.9.8 Optimization

The main problem with optimization is that of computer time. It takes a lot of it
when running an optimization on a complex model. This is related to the number of
variables, the types of variables (discrete or continuous), the number of constraints on
the solution space, the type of constraints (simple bound, linear, or nonlinear), and the
type of objective function (linear, quadratic, or nonlinear). Assume that a simulation takes
10 minutes per replication and that 5 replications are conducted at each combination of
factor values. Further, assume that the optimization algorithm calls for 1000 scenarios
(not unusual). To obtain a near-optimal solution would take 50,000 minutes. That is
approximately one month of elapsed time. Decision makes usually do not have the luxury
of this amount of time.

The time can be decreased by using fractional factorial experiments to reduce

the number of combinations. Another technique, currently in use by at least one software
vendor, is to distribute the simulation runs over the computers on a local area network.
For example, if there are ten computers available on your network, roughly ten times as
many runs can be made overnight or over the weekend. The results are fed back to the
user’s computer.
 53

As far as simulation research is concerned, optimization is an area that will continue to

receive attention.

REFERECNES

Banks, J., J. S. CARSON, AND J. N. SY [1995], Getting Started with GPSS/H, 2d ed.,
Wolverine Software Corporation, Annandale, VA.
BANKS, J. [1996], “Interpreting Software Checklists,” OR/MS Today, June.
BANKS, J. [1998], “The Future of Simulation Software: A Panel Discussion,” Proceedings
of the 1998 Winter Simulation Conference, D. J. Medeiros, E. Watson, J. Carson,
S. Manivannan, eds., Washington, DC, Dec. 13-16.
BANKS, J. [2000], “The Future of Simulation,” Proceedings of the European Simulation
Multi-Conference, Ghent, Belgium, May 23-26, to appear.


 54

Statistical Models in Simulation

5
Review of terminology and Concepts
1. Discrete random variable: Let X be a r, v. If the no. of possible values of is finite.
Or countable infinite, X is called a discrete r. v.

Note: Let X be a discrete r, v, P(X) = P (X=x) gives the probability that

r, v, equals the value of x, The collection of pairs (xi, P(xi), I = 1,2,… is
called the probability distribution of X, and P (Xi) is called the
probability mass function (p.m.l.) of X. The Possible values of X are
given by the range space of X, denoted by Rx.

2. Continuous r,v. : If the range space Rx of the random variable X is

an interval or collection of intervals, X is called a continuous r.v. for
a continuous r, v, X. the probability that X lies in the interval [a, b,]
is given by.

P (a < X < b) = f(x) dx

The function, f (x) is called the p.d.f. of the r.v.X.

Proportion of p.d.f.

(i) f(x) > 0 Ax E Rx (ii)  f ( x)dx  1

(iii) P(X=x) =0, if X is continuous r.v.

(iv) P(a<X<b)=P(a<X<b)= P(a<X<b)

3. Cumulative distribution function (edl) : The edf, denoted by F(x),

measures the probability that the r,v,X assumes value less than or
equal to x. That is
 55

F(x) = P (X < x).

If X is discrete, then F(x) =  P (xi)

If X is continuous, then F(x) = f (x) dx

Proportion of cdf :

(i) if a < b, F(a) < F(b)

(ii) lim F(x) = 1. Lim F(x) = 0

(iii) for discrete r. v. X.

P (a< X < b) = F(b) –F (a)

(iv) For continuous r. v. X.

P(a< X < b) = P (a< X < b) = F (b) – F (a).

4. Expectation and variance :

If X is a r, v., the expected value of X, denoted by E (X), is given by E

(X) = E x.P(x)

If x is discrete

x.f(x) =  x.p(x) if x is discrete

=  x.f(x)dx if x is continuous

E (X) is also called mean of X. It is also denoted by .

The variance of r. v. x. denoted by v(x) or o2, is defined by

V(X) = E[X-E(X)]2

= E (X2) – (E(X))2
 56

V(X) is the measure of spread or variation of the possible values of X

around mean E(X). Standard deviation = s.d = V ( x)  0

Mode: In the desecrate case, the mode is the value of the r. v. that
occurs most frequently. In the continuous case, the mode is the value
at which the pdf is maximized

Discrete Distribution

1. Bernoulli distribution let X = 1 if a Bernoulli trial results in success. X >

O if a Bernoulli trial results in a failure.

P(X =1) = p and P (X = 0)= q

E (X) = 1.p + 0.q = p

V (X) = [12 .p + o2.q] – p2 = p – p2 =pq

2. Binomial distribution :

The r. v. X that denotes the no, of success in n Bernoulli trials has a

Binomial distribution given by

P(x) = (n/x) Px qn-x x = 0,1,2…., n.

We have, E(X) = np, V(X) = npq.

Note : Let X1, X2, …, Xn be n index. Berroulli r. vs each with mean p

and variance pq. Then X = X1+ X2+….Xn follows Binomial distribution.

3. Geometric Distribution :

Let X be the no. Of trials requital to achieve the first success in a

sequence of Bernoulli trials.

: P (x) = P (X=x) = q x-1.p x = 1,2…

 57

E (x) = 1/p, V(X) = q/p2

4. Poisson Distribution: If the no. Of Bernoulli trials becomes very large (n

– oo) and probability of success becomes very small (p-0) then X the
no, of successes follows Poisson distribution. We have

P (x) = e-x/x!’ x = 0,1,2,…

 = E(x) =  = V(x).
Graded Questions

1. An industrial chemical that will retard the spread of fire in paint has been developed.
The local sales representative has determined, from past experience, that 48 % of
the sales calls will result in an order.

a) What is the probability that the first order will come on the fourth
sales call of the day ?

b) If eight sales calls are made in a day, what is the probability of

receiving exactly six orders ?

c) If four sales calls are made before lunch, what is the probability that
one or less results in an order ?

2. The Hawks are currently winning 0.55 of their games. There are 5
games in the next two weeks. What is the probability that they will win
more games than they lose ?

3. For the random variables, X1 and X2, which are exponentially

distributed with parameter L =1, compute P(X1 +X2 > 2).

4. Arrivals at a bank teller’s cage are Poisson distributed at the rate of 1.2
per minute.

(a) What is the probability of zero arrivals in the next minute ?

(b) What is the probability of zero arrivals in the next 2 Minutes ?

 58

5. Records indicate that 1.8 % of the entering students at a large state

university drop out of school by midterm. Using the Poisson
approximation, what is the probability that three or fewer students will
drop out of a random group of 200 entering students ?

6. The lifetime, in years of a satellite placed in orbit is given by the

following pdf :

F(x) = 0.4e –0.4x x>0

0 Otherwise

(a) What is the probability that this satellite is still “alive” after 5 years ?

(b) What is the probability that this satellite dies between 3 and 6 years
from the time it is placed in orbit ?

7. (The Poisson distributior ; can be used to approximate the binomial

distribution when n is large and p is small , say p less than 0.1 in
utilizing the Poisson approximation, Let L = np.) In the production of
ball bearings bubbles or depressions occur rendering the ball bearing
unfit for sale. It has been determined that, on the average, one in every
800 of the ball bearing has one or more of these defects. What is the
probability that a random sample of 4000 will yield fewer than three ball
bearing with bubbles or depressions ?

8. for an exponentially distributed random variable, X, find the value of L

that satisfies the following relationship :

P (X  3) = 0.9 P (X  4)

9. Accident at an industrial site occurs one at a time, independently, and

completely at random at a mean rate of one per week. What is the
probability that no accidents occurs in the next three weeks ?

10. Suppose that a Die-Hardly Ever battery has an exponential time-to-

failure distribution with a mean of 48 months. At 60 manths, the battery
is still operation.
 59

11. The daily use of water, in thousands of liters, at the hardscrabble Tool
and Die works follows a gamma distribution with a shape parameter of
2 and a scale parameter of ¼ . what is the probability that the demand
exceeds 4000 liters on any given day?

12. When Admiral Byrd went to the north Pole, he wore battery-powered
thermal wear. The batteries failed instantaneously rather than
gradually. The batteries had a life that was exponentially distributed
with a mean of 12 days. tHe trip took 30 days, Admiral Byrd packed
three batteries, what is the probability that three batteries would be a
sufficient number to keep the Admiral warm?

13. Find the probability that 6<X<8 for each, of the distribution :

Normal (10,4)

Triangular (4,10,16)

Uniform (4,16)

14. Professor Dipsy Doodle gives six problems on each exam. Each
problem requries an average of 30 minutes grading time for the entire
class of 15 students. The grading time for each problem is
exponentially distributed and the problems are independent of each
year.

(a) What is the probability that the Professor will finish the grading in 2
½ hours or less?

(b) What is the most likely grading time?

(c) What is the expected grading time?

1.5 Three shafts are made and assembled in a linkage. The length of
each shaft, in centimeters, is distributed as follows:

Shaft 1: N(60, 0.09)

Shaft 2: N(40,0.05)
 60

Shaft3 : N(50, 0.11)

(a) what is the distribution of the linkage?

(b) What is the probability that the linkage will be longer than
150.2 centimeters?

(c) The tolerance limit for the assembly are (149.83, 150.21).
What proportion of assemblies are within the tolerance
limits? [Hint : If (X1) are n independent normal Random
variables, and if Xi has mean and variance o2I, then the
sum

Y =X1 +X2+…+Xn

  i and variance   i2 ]
n n
Is normal with mean i 1 i 1

15. The time to failure of a nickel –cadmium battery is Weibull

distributed with parameter, v = 0,

B = ¼ , and a = ½ years,

(a) Find the fraction of batteries that are expected to fail prior to 1.5
years.

(b) What fraction of batteries are expected to last longer than the
mean life?

(c) What fraction of batteries are expected to fail between 1.5 and
2.5 years?

16. Explain in detail the 3-step approach of Naylor and finger in the
validation process.
AREA UNDER STANDARD NORMAL CURVE :

Z 0.0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
0.0 0.0000 0.0040 0.0080 0.0120 0.0160 0.0199 0.0239 0.0279 0.0319 0.0359
 61

0.1 0.040 0.0438 0.0478 0.0517 0.0557 0.0596 0.0638 0.0675 0.0714 0.7530
0.2 0.0783 0.0832 0.0871 0.0910 0.0948 0.9870 0.1026 0.1064 0.1103 0.1141
0.3 0.1179 0.1217 0.1255 0.1293 0.1331 0.1368 0.1406 0.1443 0.1480 0.1517
0.4 0.1554 0.1591 0.1628 0.1664 0.1700 0.1736 0.1772 0.1808 0.1844 0.1879
0.5 0.1915 0.1950 0.1985 0.2019 0.2054 0.2088 0.2123 0.2157 0.2190 0.2224

0.6 0.2257 0.2291 0.2324 0.2357 0.2389 0.2422 0.2454 0.2486 0.2517 0.2549
0.7 0.2580 0.2611 0.2642 0.2673 0.2704 0.2734 0.2764 0.2794 0.2823 0.2852
0.8 0.2881 0.2910 0.2939 0.2967 0.2995 0.3023 0.3051 0.3078 0.3106 0.3133
0.9 0.3159 0.3186 0.3212 0.3238 0.3264 0.3289 0.3315 0.3340 0.3365 0.3389
1.0 0.3413 0.3438 0.3461 0.3485 0.3508 0.3531 0.3554 0.3577 0.3599 0.3621

1.1 0.3643 0.3685 0.3686 0.3708 0.3729 0.3749 0.3770 0.3790 0.3810 0.3830
1.2 0.3849 0.3869 0.3888 0.3907 0.3925 0.3944 0.3962 0.3980 0.3997 0.4015
1.3 0.4032 0.4049 0.4066 0.4082 0.4099 0.4115 0.4131 0.4147 0.4162 0.4177
1.4 0.4192 0.4207 0.4222 0.4236 0.4251 0.4265 0.4279 0.4292 0.4306 0.4319
1.5 0.4332 0.4345 0.4357 0.4370 0.4382 0.4394 0.4406 0.4418 0.4429 0.4441

1.6 0.4452 0.4463 0.4474 0.4484 0.4495 0.4505 0.4515 0.4525 0.4535 0.4545
1.7 0.4554 0.4564 0.4573 0.4582 0.4591 0.4599 0.4686 0.4616 0.4625 0.4633
1.8 0.4641 0.4649 0.4656 0.4664 0.4671 0.4678 0.4686 0.4693 0.4699 0.4706
1.9 0.4713 0.4719 0.4726 0.4732 0.4738 0.4744 0.475 0.4756 0.4761 0.4757
2.0 0.4772 0.4778 0.4783 0.4788 0.4793 0.4798 0.4803 0.4808 0.4812 0.4817

2.1 0.4821 0.4826 0.4830 0.4834 0.4838 0.4842 0.4846 0.4850 0.4854 0.4857
2.2 0.4861 0.4864 0.4868 0.4871 0.4875 0.4878 0.4881 0.4884 0.4887 0.4890
2.3 0.4893 0.4896 0.4898 0.4901 0.4904 0.4906 0.4909 0.4911 0.4913 0.4916
2.4 0.4918 0.4920 0.4922 0.4925 0.4927 0.4929 0.4931 0.4932 0.4934 0.4936
2.5 0.4938 0.4940 0.4941 0.4943 0.4945 0.4946 0.4948 0.4949 0.4951 0.4952

2.6 0.4953 0.4955 0.4956 0.4957 0.4959 0.4960 0.4961 0.4962 0.4963 0.4964
2.7 0.4965 0.4966 0.4967 0.4968 0.4969 0.4970 0.4971 0.4972 0.4973 0.4974
2.8 0.4974 0.4975 0.4976 0.4977 0.4977 0.4978 0.4979 0.4979 0.4980 0.4981
2.9 0.4981 0.4982 0.4982 0.4983 0.4984 0.4984 0.4985 0.4985 0.4986 0.4986
3.0 0.4987 0.4987 0.4987 0.4988 0.4988 0.4989 0.4989 0.4989 0.4990 0.4990
STATISTICAL MODELS IN SIMULATION
 62

In real world phenomena these are few setuations where the actions of
the entities with in the system under study can be completely predicted in
advance.

An appropriate model can be developed by considering the samples

of interest.

Terminology & concepts.

1. discus random variables : let us consider X as a random variable. Of

the no of values of X is finite or countably inficiate then X is called a
discrete random variable.

Let X be a discrete random variable with each possible outcome x, in

Rx then P(x i) = P(x = xi)Gives the probability that the random variable
equals the value of xi.

The no. P(xi), I=1,2… must satisfy the following two conditions.

1) P (Xi)  0 I

 p( x )  1

2) i

i 1

The collection of pair (Xi, P (Xi) ), I = 1,2…, is called the probability

distribution of X. & P (xi) is called the Probability more function (pmf) of X.

2. Continuous random variables: - Of the range space RX of the

random variables X is an interest or a collection of intervals X is an interval
or a collection of intervals, X is called a continuous random variable.

For a continuous random variable X, the probability that X lies in the

interval [a, b,] is given by –

 f ( x)dx ------------------------------------------(1)
b
P (a  X  b) =
a
 63

The function f (x) is called the probability density function (pdf) of the
random variable X. the pdf satisfies the following conditions.

(a) f(x) > 0 for all x in Rx.

(b) Rx f(x) dx = 1

(c) F (x) = 0 if x is not in Rx.

As a result eq n (1) for any specified value xo

P (x = Xo) = Oxo

 f ( x)dx =0
x0

x0

.: f (x) dx = O

.: P ( X = Xo) = o, the following eqn becomes

P(a X  b) = P (a < x  b) = P (a  x < b) = P (a  x < b) ----------

(2)

it can be graphically represented as :-

f(x)

x=a x=b x

the shaded area represents the probability that X lies in the interval [a, b]
 64

3) Cumulative distribution function : the cumulative distribution

function (Cdf), denoted by F(x) measures the probability that the
random variable X assumes a value less than or equal is X. that
is F(x) = P (x < x)

Of X is discrete then
F (x) =  p (xi) ------------------------------------------------ (3)
xi x

Of X is continuous then

 f (t )dt
x
F (x) = f(x) dt ---------------------------------(4)


Some properties of the cdf are :-

i) F is a nondecresing f n of a<b then F(a) F(b).

ii) it F (x) = 1

iii) it f(x) = o

the probability of X can be defined using cdf

P (a< x  b) = f (b) – f (a) , a < b ------------------------------ (5)

If X is a discrete random variable with possible values x1, x2,where

x1<x2<…,the cdf is a stepfth. The value of the cdf is constant in the
interval [xi-1,xi] and take the size of p (xi) at xi.

(4) Expectation: - An important concept in probability theory is that of

the expectation of a random variable of x is a random variable the
expected value of x denoted by E (X), for discrete and continuous
variables is defined as follows :

E(X) = XiP (xi) is X is discrete & E (X) = 

i
xif(xi) dx if X is discrete.------

----(6)
 65


E(X) = XiP (xi) is X is discrete & E (X) =  xf(x) dx

if X is continuous ------

------(7)

The expected value E(x) of a random variable X is also referred to as the

mean X, or the first moment of X. the quantity E (Xn), n>1. Is called the
Nth moment of X & is computed as follows :

E (Xn) = 

i
Xn P (Xi) if X is discrete ----------------------(8)


& E (Xn) =  X
n
f(x) dx if X is Continues ------------(9)


the variance of a random variable, X, denoted by V(x) or var ( X) or is

defined by

V(X) = E [X –E(x) 2]

A useful identify in computing V(X) is given by

V(X)=E (X2) – [ E(X) 2 ] ----------------------------------- (10)

Where E(x) is central tendency of a random variable.

The variance of x measures the expected value of the rquared

difference between the random variable & its mean.

V(x) is a variation of the possible values of X around the mean E(x).

The standard deviation, is defined to be the rquare root of the

variance , o2 the mean E(X) & std deviation   v(x) -------------------------
-(11) is defined by eqn. (11)
 66

(5) The Mode :- In the discrete case, the mode is the value of the random
variable that occurs most frequently, in the continuous case, the mode is
the value of which the pdf is marimized. The mode may not be unique. Of
the modal value occurs at two values of the random variable, the
distribution is said to be bimodal.

Useful statistical models

In a queucing system, interarrival and service times are often probabilitic.
In an inventory model, the time between demand and had times may be probabilitic .in a
reliability model, the time to failure may be probabilistic. In each of these instances the
simulation analyst desires to generale random extents and to use a known statistical
model if the underlying distribution can be found.

1) Queuing Systems: - In the queuing examples, interarrival and

service times are given. The times between arrivals and service times
were always probabilistic time or a constant service time.
Service times may be constant or probabilitic. Of service times are
completely randam, the expenetial distribution is often used for simulation purpose.
But these are several possibilities. At may be possible that the service times are
constant but some random variability causes fluctuations in either a positive or
negative way.

The gamma and weibull distibutions are also used to model

interarrival and service times. The difference between the experential.
Gamma and welbull distributions involve the location of the models of the
pdf’s and shapes of these fails for large & small times. The experential
distribution has its mode at the origin but the gamma & weibull
distributions have there modes at some pt (>o) which is a function of the
parameter values selected.

2) Inventory Systems :- In realistic inventory systems there are three

random variables:
i) the no of units demanded per order or per line period.

ii) The time between demands:

iii) The lead time

 67

In very simple mathematical models of inventory systems demand

is a constant over time & lead time is zero or a constant.

Hence, in simulation models, demand occurs randomly in time.

The geometric distribution, which is a special case of the –ve

benomical, has its mode at unity, given that affeart one demand has
occurred. Of demand data are characterized by a long, the –ve binomical
distribution may be appropriate.

The poison distribution is often used to model demand becouse it is

simple, it is extensively tabulated & it is well known. The fail of the passion
distribution is generally shorter than that of the negative binomial. Is for
the same set, fewer large demands will occur if a poisson model is used
than if a negative binomial distribution is used.
3) Reliability and Maintainability :- Time to failure has been
modeled with numerous distributions. Including the expenential, gamma &
weibull. If only random failures occur, the time to failure distribution may
be modeled as expenential. The gamma distribution arises from modeling
standlay redundancy where each component has an expenential time to
failure. The weibull distribution has been extensively used to represent
time to failure & he nature is such that it may be made to approcimate
many obsorved phenomena.
4) Limited Data :- In many instances simulations begin before data
collection has been completed. Three distribution have application to in
complete or limited data. These are the uniform, triangular & beta
distributions. The uniform distribution can be used when an interarrival or
service time is known to be random, but no information is immediately
available about the distribution the triangular distribution can be used
when atsumptions are mode about the minimum, maximum & modal
values of the random variable. Beta distribution provides a variety of
distributional farms an the unit intervals
5) Other Distributions :- Several other distributions may be useful in
discrete system simulation. The bernouli & binomial distributions are two
discrete distributions which may describe phenomena of interest.
 68

DISCRETE DISTRIBUTIONS
Discrete random variables are used to describe random phenomena in
which only integer values can occur.

1. Bernoulli trials and the Bernoulli Distribution :- Consider in trials

each of which can be a success or a failure.
th
Let Xj =1 if j is a Success.
th
Xj = 0 if j is a failure.
The n Bernoulli trials are called a Bernoulli process if the trials are independent. Trial
has only two possible outcomes is success or failure.

There fore, probability can be defined as:- P (x1, x2 … xn) = P1 (x1). P2 (x2)…
Pn (xn) & Pj (Xj) = P (Xj) = { P,
Xj = 1, j = 1,2 …n
{ 1-P = q, Xj = 0 , j = 1,2…n
{ O, othervise )

----------------(11)

for one trial, given (eq n 11) distribution is called the bernoulli distributions.
: Means & variance of Xj are defined as –
E(Xj) = 0.q + 1.p = p
& V(Xj) = [ (O2.q) + (12.P) ] – P2
= P (1-P)

BINOMIAL DISTRIBUTION :- The random variable X denotes the no of

success in n Bernoulli trials has a binomial distribution given by P(X) where

P(x) = ( nx ) pxqn-x , x=0,1,2…..n

, otherwise -------------------------------------(12)

The probability of eq n (12) in calculated as :-

P(SSS…SS FF…FF ) = Px qn-x

x values n-x values

 69

Where q = 1-p

The Bernoulli random variables X with mean P & variance P (1-P) = pq

can be expressed as –

X = X1 + X2+….+ Xn

: mean E (X) = P + P+ … +P = np --------------------------------13)

Variance V(x) = pq + pq+…+pq = npq ------------------------- (14)

GEOMETRIC DISTRIBUTION :- This distribution is related to a sequerce

of bernoulli trials, the Random variable of interest X is defined to be the no
of trials to achieve the first success.
x-1
: . P(x) = { q p , x=1, 2…
{ 0, otherwise --------------------------------------------------(15)

2
The event { X=x } occurs when these are x-1 failure followed by a success.
Probability of failure = q = 1-p
Probability of success = p
P(FFF…..FS) = q
x-1
p
Mean = E(X) = 1/p ---------------------------------------------------------------(16)

2
& Variance = V(x) = q/p --------------------------------------------------------(17)

POISSON DISTRIBUTION :- The poisson probability mass f n is given by

 e   x
- P(x) =  , x=0,1,…… ------
 x!
-(18)

Generally poisson distribution descrebes random processes & generally it

is simple sample. Where a < o.

mean = E (X) = x = V(x)

and cumulative distribution f n is given by
 70

e   i
F ( x)  
x
------------------------------(19)
i o i!

Continuous Distributions :- Continuous random variables can be used

to discribe random phenomena in which the variable of interest can take on
any value in same interval. (I)

Uniform Distribution :- A random variable X is uniformaly distributed on

the interest (a,b) if its pdf is given by –

F(X) = { 1/b-a, a  x  b _____(20)

{o, otherwise

The cdf is given by -

f(x) = { o, x  a
{ x-a/x-a, a  x  b ________(21)
{1, xb
x 2  x1
 P (X1 < X < X2) = F(X2) – F (X1) =
ba

Which is propertional to the length of the interval for all X1 & X2 satisfying a
< X1 <X2 < b. :. Mean & variance can be defined as .

ab
E(X) = --------------------------------------------------------(22)
2

( a  b) 2
V(X) = ----------------------------------------------------(23)
12

(ii) EXPENENTIAL DISTRIBUTION :- A random variable X is said to be

expenentially distributed with parameter X > O if its pdf is given by -
f (X) = { e  x , x  o ------------------------------
---(24) { o, else
where
 71

The exponentical distribution has been used to model intervival

times when arrivals are completely random and to model service times
which are highly variable.
In these instances is a sate; arrivals per hr service per min.

The exponential distribution has mean & variance given by

E(X) = 1/ & V(X) = 1/ --------------------------------- (25)

Thus the mean and standard deviation are equal.  Cdf can be expressed
as –

F(X) = { o, X<O ------------------------------------(26)

{   t dt  1  e x , x 
0

(iii) Gamma Distribution A function used in defining the gamma

distribution is the gamma function which is defined for all  > o as

 () = x
 1  x
e dx -------------------------(27)
0
by integrating equation (27) it became

 () = (-1) (-1) ------------------------------(28)

If  is an integer, then using  (1) = 1, equation (28) becomes,

 () = (-1) ! ---------------------------------------(29)

The gamma Fn can be thought of as generalization of the factorial notion
which applies to all the nos. not just integers.

A random variable X is gamma distributed with parameters  &  if its pdf is given by

 
 (  )  1 e   ,
 
f(x) = X>O
 ( )
 72

{ O, otherwise ------------------------(30)

The parameter  is called the shape parameter and  is called the scale
parameter.

The mean and variance of the gamma distribution are given by –

E(x) = 1/ --------------------------------------------------------(31)

& V(x) = 1/B -------------------------------------------------------------------------------(32)

The cdf of X is given by -

  
F(x) = 1   ( t )  1 e  t dt , x>0
   (  )

{ 0, X  0 -------(33)

When  is an integer, the gamma distribution is related to the exponential distribution in

the following manner :

If the random variable, X is the sum of B independent, exponentially distributed random

variables, each with parameter BQ, then X has a gamma distribution with parameters B &

X = X1 + X2+ …. + X  --------------------------------------------------(34)
 73

where the pdf of Xj is given by -

g ( Xj )  (  )e x 
xj
,

{ 0, otherwise

and Xj are matually independent.

(iv) Erlang Distribution :- The pdf of equation,

 
f ( x)   (  )  1 e  x , x > 0
 (  )

{ 0, otherwise

is often referred to as the Erlang distribution of order K when B = K, an integer.

The mean & variance of gamma distribution can be used to define the mean & variance

but regardless of the value of B.

Ie. E(X) = 1/ & V(X) = 1/ 

2

if  = k, an integer

 Mean of the distribution can be difined as –

E(X) = E(X1) + E(X2) + …. + E(Xk)

The expected values of the exponentially distributed Xj are each given by 1/k.
 74

Thus,

E( X )    ......  
1 1 1 1
K K K 

If the random variables Xj are independent the variance of their sum is the sum of the

variance or

V (X )    ......  
1 1 1 1
( K ) 2
( K ) 2
( K ) 2
K 2

when B=k , a +ve integer, the cdf can be defined as -

 k 1 e0 kx ( kx )i
F ( x)  1   , x>0 -------------------------------------(35)
 i 0 i!

Which is the sum of poisson terms with mean  = Kx.


 75

6
Queuing Models
 Characteristics of Queuing systems :

The key elements of queuing systems are the customes and servers. The
term customer can refer to anything that arrives at a facility and requires
service (e.g. people, machines, patienes, e-mail etc.) the term “server”
might refer to any resource which provides the requested service. (e.g.
receptionist, mechanics, medical personnel, CPU in a computer etc.)
1. The calling population :-

The population of potential customers, referred to as the calling

population, may be finite or infinite. Examples of infinite population may be
finite but large we may use infinite population models provided that the
number of customers being served or waiting for service at any given time
is a small proportion of the population of potential customers.
As an example of finite population
consider, a bank of five machines which may break down and single
mechanic is available for service.

In an infinite population model, the arrival rate is not affected by the

number of customers who have left the calling population and joined the
queuing system. For the finite calling population models, the arrival rate to
the queuing system does depend on the number of customers being
served and waiting.
2. System capacity :-

In many queuing systems there is a limit to the number of

customers that may be in the waiting line or systems. An arriving customer
 76

who finds the system full does not enter but returns immediately to the
calling population e.g. waiting room of a clinic.
Some systems may be considered as having unlimited capacity.
E.g. Ticket counters at a railway station.

3. The arrival process :-

The arrival process for the models is characterized in terms of

interarrival times of successive customers. Arrivals may occur at
scheduled times or at random times. When arrivals are at random, the
interararrival times are usually characterized by a probability distribution.

The most important model for random arrivals is the Poisson arrival
process. For a Poisson arrival rate is customer per unit time. If An
represent interarrival time between customer n-1 and n, then for a Poisson
arrival process, An is exponentially distributed with mean 1/ time units.

4. Queue Behavior and Queue Discipline :-

Queue behavior refers to customer actions while in a queue waiting

for service to begin. There is possibility that incoming customers may balk
(i.e. leave if the line is too long ), renege (i.e. leave after being in the line if
the line is moving too slowly), or jockey (move from one line to another),
Queue discipline refers to the logical ordering of customers in a queue,
common queue disciplines are first-in, first-out (FIFO); last-in, first-out
(LIFO); service in random order (SIRO); Shortest processing time first
(SPT); and service according to priority (PR).

5. Service Times and Service Mechanism :-

The service times of successive arrivals are denoted S1, S2, S3,….
They may be constant or of random duration. If service times are random,
then { S1, S2,…} is usually a sequence of independent and identically
distributed random variables. The exponential. Erlang, Weibull etc. are
 77

used to represent service times in different situations.

 Queuing Notation :-

Kendall [1953] proposed a notational system for parallel server

systems. It is in the format A/B/C/N/K.
A : the interarrival – time distribution
B : the sercice – time distribution
C : the number of parallel servers

N : system capacity

K : the size of the calling population.

Common symbols for A and B are :
M : Exponential or Markov
D : constant or deterministic

Ek : arbitary or general
GI : general independent.

Let
Pn : Steady-state probability of having n customers in system.
Pn(t): Probability of n customers in system at time t.
L : Arrival rate
Le : Effective arrival rate
U : Service rate of one server
P : Server utilization
An : Interarrival time between customers n-1 and n
Sn : Service time of the nth arriving customer.
Wn : Total time spent in system by the nth arriving customer.
Wnq : Total time spent in the waiting line by customer n
L (t) : The number of customers in system at time t.
Lq(t) : The number of customers in queue at time t.
L : long-run time-average number of custmers in system.
Lq : Long-run time-average number of customers in queue.
 78

W : Long-run average time spent in system per customer.

Wq : Long-run average time spent in queue per customer.
Note :

(I) All systems will be assumed to have a FIFO queue discinline.

(II) The term system usually refers to the waiting line plus the service
mechanism; whereas the term ‘queue’ refers to the waiting line
alone.

 Long-run Measures of performance of Queuing system.

1. Time-Average number in System, L :

Consider, a queuing system over a period of time T.

Let L (t) denote the number of customers in the system at time t.
Let Ti denote the total time during [o,T,] in which the system contained
exactly I customers. In

 Ti  T

General,
i 0
The time-weighted-average number in the system is defined by

 Ti 
 
1  
L  i  ………………………… (1)
^
i.Ti
T i0 i0  T 

 i.Ti   L(t )dt

 T
We Have,
i 0 O

 L   L(t )dt ………………………………. (2)

1T^

TO

The expression in equations (1) and (2) are always equal for any queuing system,
regardless of the number of servers, the queue discipline, or any other special
circumstances. This average is also called time-integrated average.
 79

As time T gets large, the observed value L approaches a limiting value,

say L, which is called the long-run time-average number in the system.
That is

T O
T
L L(t )dt  LasT   ……………….. (3)
^ 1

Equation (1)and (2) can also be applied to any subsystem of queuing

system.
Let Lq(t) denote the number of customers waiting in line, and Tiq denote
the total time during [0,T] in which exactly I customers are waiting in line,
then

 iTi q   Lq (t )dt …………………. (4)

T
1 q
Lq 
^ 1
T i 0 TO

L q  LqasT   with probability 1 ….(5)

^
And

Where Lq is the observed time-average number of customers waiting in line from time 0
to T and Lq is the long-run time-average number of customers waiting in line.

2. Average Time spent in system per customer, w;

Consider, a queuing system over a period of time T.

Let Wi be the time spent by customer I in the system during [0, T]. The
average time spent in the system per customer, called the average system
time, is given by

W ……………………………………….(6)
N
wq 
^
1
i
N i 1

Where N is the number of arrivals in [0, T].

 80

Let Wiq denote the total time customer I spends waiting in queue. Let Wnq
be the observed average time spend in the queue(called delay) and Wq
be the long-run average delay per customer. Then

W ……………………………… (7)
N
wq 
^
1 q
i
N i 1

w q  wq asN   ……………. (8)

^
And

Example :

Consider a simulation of queuing system as shown below, Obtain L, Lq and Wq

Customer Time of Time of

Arrival Departur
e
1 0 2
2 3 8
3 5 10
4 7 14
5 16 20

Graded Questions

1. A two-runway (one runway for landing, one for taking off) airport is being designed for
propeller-driven aircraft. The time to land an airplane is known to be exponentially
Distributed with a mean of 1 ½ minutes. If airplane arrivals are
assumed to occur at random.
What arrival rate can be tolerated if the average wait in the sky is not to
exceed 3 minutes?

2. The port of Trop can service only one ship at a time. However, there is
mooring space for three more ships. Trop is a favorite port of call, but if
 81

no mooring space is available, the ships have to go to the Port of

Poop. An average of secen ships arrive each week, according to a
Poisson Process. The port of Trop has the capacity to handle an
average of eight ships a week, with service times exponentially
distributed. What is the expected number of ships waiting or in service
at the Port of Trop?

3. At Tony and Cleo’s bakery, one kind of birthday cake is oftered. It

takes 15 minutes to decorate this particular cake, and the job is
performed by one particular baker, in fact, this is all this baker does,
what mean time between arrivals (exponentially distributed) can be
accepted if the mean length of the queue for decorating is not to
exceed five cakes?

4. Patients arrive for a physical examination according to a Poisson

process at the rate of one per hour. The physical examination requires
three stages, each one independently and exponentially distributed
with a service time of 15 minutes. A patient must go through all three
stages before the next patient is admitted to the treatment facility.
Determine the average number of delayed patients, Lq for this system.
[Hint : The variance of the sum of independent random variables is the
sum of variance.

5. Suppose that mechanics arrive randomly at a tool crib according to a

Poisson process with rate L = 10 per hour. It is known that the single
tool clerk services a mechanic in 4 minutes on the average. With a
standard deviation of approximately 2 minutes. Suppose that
mechanics make $15 per hour. Determine the steady-state average
cost per hour of mechanics waiting for tools.

6. Arrivals to an airport are all directed to the same runway. At a certain

time of the day, these arrivals form a Poisson process with rate 30 per
hour. The time to land an aircraft is a constant 90 seconds. Determine
Lq, (I), L and (I) for this airport. If a delayed aircraft burns $5000 worth
of fuel per hour on the average, determine the average cost per aircraft
of delay waiting to land.
 82

7. A machine shop repairs small electric motors which arrive according to

a Poisson process at a rate of 12 per week (5-day, 40-hours work
week). An analysis of past data indicates that engines can be repaired
or an average, in 2.5 hours, with a variance of 1 hour2, I low many
working hours should a customer expect to leave a motor at the repair
shop (not krowing the status of the system)? If the variance of the
repair time could be controlled, what variance would reduce the
expected waiting time to 6.5 hours?

8. Arrivals to a self-service gasoline pump occur at a Poisson fashion at a

rate of 12per hour. Service time has a distribution which averages 4
minutes with a standard deviation of 11/3 minutes. What is the
expected number of vehicles in the system?

9. A large customer shopping mall is to be constructed. During busy time

the arrival rate of cars is expected to be 1000 per hour. And based on
studies at other malls, it is believed that customers will spend 3 hours,
on average, shopping, the mall designers would like to have sufficient
parking so that there enough spaces 99.9% of the time. How spaces
should they have?

10. A small copy shop has a self-service copier. Currently there is room for
only 4 people to line up for the machine (including the person using the
machine); when there are more than 4 people, then the additional
people must line up outside the shop as much as possible. For that
reason they are thinking about a second self-service copier. Self-
service customers have been observed to arrive at a rate of 24 per
hour, and they use the machine 2 minutes, on average. Assess the
impact of adding another copier. Carefully state any assumption or
approximations you make.

[Note : For Q.11 to Q16, Assume exponential service time]

11. Customers arrive at a one man barber shop according to a Poisson process with a
mean interarrival time of 20min. Customers spend an average of 15 min in the
barber’s chair,
 83

(a) what is the expected number of customers in the barber shop? In the
queue?

(b) What is the probability that a customers in the barber shop? In the
queue?

(c) How much can a customer expect to spend in the barber shop?

(d) Management will put another chair and hire another barber when a
customer’s average waiting time in the shop exceeds 1.25 h. how
much must the average rate of arrivals increase to warrant a second
barber?

(e) What is average time customers spend in the queue?

(f) Obtain the probability that waiting time in system is greater than 30
minutes

(g) Obtain the probability that there are more than three customers in the
system.

Ans :- [3, 2.25; 0.25; 1h; 3.2phr; 0.75h; 0.61; 0.32]

12. If customers arrive for service according to a Poisson distribution at the

average rate of 5 per day, how fast must they be serviced on the
average (assume exponential service time) in order to keep the
average number of customers in the system less than 4?
Ans :- [3.84
h/service]

13. If the mean arrival rate is 24 per hour, find from the customer’s point of
view of the time spent in the system, whether 3 channels in parallel
with a mean service rate of 10 per hour is better or worse than a single
channel with mean service rate of 30 per hour.
Ans :- [single
channel is
better.]
 84

7
Random Number Generation

Random numbers are necessary basic ingredient in the simulation of

almost all discrete system. Most computer languages have a subroutine or
function that will generate a random number.

1. Properties of Random Numbers.

A sequence of random number R1, R2,…, must have two statistical properties,
uniformity, and independence. Each random number R1 is an independent sample drawn
from a continuous uniform distribution between zero and 1. That is, the pdf is given by.

F(x) = 1, 0 < x < 1

0, otherwise

The expected value of each Ri is given by

E ( R)  
1 2

xdx  x
1

O
2 0

1
=
2
and the variance is given by,

V ( R )   x 2 dx  E ( R )
1
2

1
3 1 2

  
x
3 0 2

  
1 1 1
3 4 12
 85

Note :

1. If the interval (0,1) is divided into n classes, or subinterval of equal length, the
expected number of observations in each interval is N/n, where N is the total no. of
observation.

2. The probability of observing a value in a particular interval is

independent of the previous value drawn.

2. Generation of Pseudo-Random Number

The word “pseudo” is used to imply that the very act of generating random numbers by a
known method removes the potential for true randomness. If the method is known the set
of random numbers can be replicated, that is the numbers are not truly random.

The goal of any generation scheme, however, is to produce a sequence of

numbers between zero and 1 which simulate, or imitates, the ideal
properties of uniform distribution and independence as closely as
possible.
When generating pseudo-random numbers certain problems or erros can
occur.

1. The generated numbers may not be uniformly distributed.

2. The generated numbers may be discrete-valued instead of continuous-

valued.

3. The mean of the generated numbers may be too high or too low,

4. The variance of the generated nos. may be too high or too low,

5. There may be dependence.

e.g.
(a) Autocorrelation between numbers.
 86

(b) Nos. successively higher or lower than adjacent nos.

(c) Several numbers above the mean followed by several nos. below
the mean.

Departures from uniformity and independence for a particular generation

scheme may be detected by statistical tests. If such departures are
detected, the generation scheme should be dropped in favour of an
acceptable generator.

14. A telephone exchange has 2 long distance operators. The telephone

company finds that during the park load, long distance calls arrives in a
Poisson fashion at an average rate of 15 per hour. The length of
service on these calls is approximately exponentially distributed with
mean length 5 min.

(a) What is the probability that a subscriber will have to wait for his long
distance call during the park hours of the day?

(b) If the subscribers will wait and are serviced in turn, what is the
expected waiting time?
Ans :- [ 0.48; 3.2
min]

15. A one-person barber shop has 6 chairs to accommodate people

waiting for a hair cut. Assume that customers who arrive when all the 6
chairs are full leave without entering the barber shop. Customer arrive
at the average rate of 3 per hour and spend an average of 15 min in
the barber’s chair.

(a) What is the probability that a customer can get directly into the
barber’s chair upon arrival?

(b) What is the expected number of customers waiting for a hair cut?

(c) How much time can a customer expect to spend in the barber
shop?
 87

(d) What fraction of potential customers are turned away?

Ans :- [ 0.2778;
1.3878; 43.8 min;
3.7%]

16. A Barber shop has 2 barbers and 3 chairs for waiting customers.
Assume that customers arrive in Poisson fashion at a rate of 5 per
hour and that each barber services customers. According to
exponential distribution with mean of 15 min. further, if a customer
arrives and there are no empty chairs in the shop, he will leave. Find
the steady-state probabilities. What is the probability that the shop Is
empty? What is the expected number of customers in the shop?

17. Two attendants manage the tool crib while mechanics, assumed to be
from an infinite calling population, arrive for service. Assume poisson
arrivals at rate of 2 meehanics per minute and exponentially distributed
service times with mean 40 seconds.

(a) Obtain the number of mechanics in the system and waiting time in
the queue.

(b) Suppose service takes on an average 40 seconds with s.d. of 30

seconds. Obtain the number of mechanics in the system.

18. What are the characteristics of queueing system and how would you
determine the costs in queueing problems.

19. Name and explain some of the useful statistical models for queuing
systems

20. What are the long run measures of performance of queuing systems?
Explain briefly.

Usually, random numbers are generated by a digital computer as a part of

the simulation. Numerous methods can be used to generate the values. In
 88

selecting among these methods. Or routines, there are a number of

important considerations:

1. The routine should be fast, individual computations are inexpensive,

but simulation may require many hundreds of thousands of random
numbers. The total cost can be managed by selecting a
computationally efficient method-number generation.

2. The routine should be portable to different computers, and ideally to

different programming languages. This is desirable so that the
simulation program produces the same results wherever it is
executed.

3. The routine should have a sufficiently long cycle. The cycle length or
period, represents the length of the random-number sequence before
previous numbers begin to repeat themselves in an earlier order. Thus,
if 10,000 events are to be generated, the period should be many times
that long.
A special case of cycling is degenerating. A routine degenerates when
the same random numbers appear repeatedly. Such an occurrence is
certainly unacceptable. This can happen rapidly with some methods.

4. The random numbers should be replicable. Given the starting point (or
conditions), it should be possible to generate the same set of random
numbers, completely independence of the system that is being
simulated. This is helpful for debugging purposes and is a mean of
facilitating comparisons between systems. For the same reasons, it
should be possible to easily specify different starting points, widely
separated, within the sequence.

5. Most important, and as indicated previously, the generated random

numbers should closely approximate the ideal statistical properties of
uniformity and independence.

3. Techniques for Generating Random Numbers

(a) Linear Congruential method

 89

The linear congruential method produces a sequence of integers

X1, X2,…. Between zero and m-1 according to the following
recursive relationship :

X I+1= (a Xi + C) mod m, I = 0,1,2…

The initial value Xo is called the seed, a called the constant
multiplier, C is the increment, and mds the modulus. If C# O, the
form is called the mixed congruential method. When C = O, the
form is known as the multiplicative congruential method. Random
numbers between zero and 1 can be generated by

R1  , i  0,1,2,.....
Xi
m

Note : The ultimate test of any generation scheme is how closely

the generated numbers R1, R2 … approximate uniformity and
independence. Consideration must also be given to maximum
density and maximum period.

Maximum density means that the values assumed by Ri, I = 1,2,…

leave no longer gaps [0,1]. This can be achieved by making m
suitably large (e.g. 231-1 or 248)

To avoid cycling (i.e. recurrence of the same sequence of generated numbers) the
generator should have the largest possible period. This can be achieved by proper
choice of a, c, m, and Xo.

(b) Combined linear congruential generators

Let Xi,1 , Xi2, …. Xi k be the ith output from k different
multiplicative congracuential generators, Let mj be the prime
modulus of the jth generator and multiplier aj is chosen so that the
period is mj-1.

 k 
Let X 1    1 X i , j  mod m1  1
j 1

 j 1 
 90

And R 1  , X1  0
X1
m1

m1  1
 , X1  0
m1

The maximum possible period for such a generator is

( m1  1)( m 2  1)......( m k  1)
P=
2 k 1
4. Test for Random Numbers

The desirable properties of random numbers are uniformity and

independence.

(1) Frequency test :-

This is a test for checking uniformity, here,

- Ho : Ri ~ U[0,1]
- H1 : Ri ~ U[0,1]

(a) The kolmogorov-simirnov test :

This test compares the continuous cdf, F(x), qf the uniform distribution
to the campirical cdf, SN(x) of the sample of N observations. We have,
F(x) = x, 0  x  1
Let R1, R2 …., RN be the sample from random number generator.
Then SN (x) is defined by

number of R1 , R2 ,...., R N S x X
SN (x) =
N
 91

As N becomes larger, SN(x) is better approximation to F(x).

Let D = max F ( x )  S N ( x)

The sampling distribution of D is known and is tabulated.

Test procedure :

Step 1: Rank the data from the smallest to largest. Let R(i) denotes the
ith smallest observation.
Let R(i)  R(2) …R(N)

Step 2 : Compute

D+ = max [i/N – R(i)]

i 1
D- = max R(1) 
N

D = max (D-, D+)

Step 3: Determine critical value, D , for the specified significance

level and the given samplesize N.

Step 4: If D > D , H 0 is rejected.

(b) The chi-square test.
Suppose the observed values are divided into n classes.
Let Oi = observed number in the ith class.
ei = expected number in the ith class.
Then,

oi  ei 2
chi  X  
n
2 2

i 1 ei
 92

N
for the uniform distribution ei =
n
d.f. = n-1

if X2 > X2 n-1, a, we reject Ho

Run : A run is defined as succession of similar events preceded and

followed by a different event.

(2) Runs test : This is a test for checking independence Ho: Ri ~

independently Hi : Ri ~ independently.

(a) Runs up and Runs down:

Here, we are concerned with number of runs. Too few or too many
runs would indicate non-random behaviour of data.

An up run is a sequence of numbers each of which is succeeded by a

large number. A down run is sequence of numbers each of which is
succeeded by a smaller number. A ‘+’ (‘_’) sign will indicate an observation
which is fallowed by a larger (smaller) number.

2N  1
Then,  = E® =
3

16 N  29
 = V(R) =
90

For N > 20, R~ N( , )

R
The test statistic is Z = ~ N(0,1)


Reject Ho if Z < -Za/2 or Z > Za/2 where  is the level of significance.

(b) Runs above and below the mean :

 93

A ‘+’ sign will be used to denote an observation above mean, and a ‘_’
sign will denote an observation below the mean.
Let, n1 = no. of observation above the mean.
n2 = no. of observation below the mean.
R = no. of runs.

Then ,  = E(R) =  ;
2n1 n 2 1
N = n1+n2
N 2

2n1 n2 (2n1 n 2  N )
  2  V ( R) 
N 2 ( N  1)

For n1 > 20 or n2 > 20, R~ N (,  )

R
Let Z =


Reject Ho if Z < -Za/2 or Z > Za/2 where  is the level of signnificance.

Colmogorov –Smirnov Critical Values

Degrees of 105.50%
Freedom D D D
(N) 0.10 0.05 0.01
1 0.95 0.975 0.995
2 0.776 0.842 0.929
3 0.642 0.708 0.828
4 0.564 0.624 0.733
5 0.51 0.565 0.669
6 0.47 0.521 0.618
7 0.438 0.486 0.577
8 0.411 0.457 0.543
9 0.388 0.432 0.514
10 0.368 0.41 0.49
11 0.352 0.391 0.468
 94

12 0.338 0.375 0.45

13 0.325 0.361 0.433
14 0.314 0.349 0.418
15 0.304 0.338 0.404
16 0.295 0.328 0.392
17 0.286 0.318 0.381
18 0.278 0.309 0.371
19 0.272 0.301 0.363
20 0.264 0.294 0.356
25 0.24 0.27 0.32
30 0.22 0.24 0.29
35 0.21 0.23 0.27
Over 1.22 1.36 1.63
35 N N N

Percentage Points of the Chi-Square Distribution with v Degrees of

Freedom

X2
 95

 20.005 20.01 20.025 20.05 20.10

1 7.88 6.63 5.02 3.84 2.71
2 10.6 9.21 7.38 5.99 4.61
3 12.84 11.34 9.35 7.81 6.25
4 14.96 13.28 11.14 9.49 7.78
5 16.7 15.1 12.8 11.1 9.2
6 18.5 16.8 14.4 12.6 10.6
7 20.3 1805 16 14.1 12
8 22.0 20.1 17.5 15.5 13.4
9 23.6 21.7 19 16.9 14.7
10 25.2 23.2 20.5 18.3 16
11 26.8 24.7 21.9 19.7 17.3
12 28.3 26.2 23.3 21 18.5
13 29.8 27.7 24.7 22.4 19.8
14 31.3 29.1 26.1 23.7 21.1
15 32.8 30.6 27.5 25 22.3
16 34.3 32 28.8 26.3 23.5
17 35.7 33.4 30.2 27.6 24.8
18 37.2 34.8 32.5 28.9 26
19 38.6 36.2 32.9 30.1 27.2
20 40.0 37.6 35.2 31.4 28.4
21 41.4 38.9 35.5 32.7 29.6
22 42.8 40.3 36.8 33.9 30.8
23 44.2 41.6 38.1 35.2 32
24 45.6 43 39.4 36.4 33.2
25 49.6 44.3 40.6 37.7 34.4
26 48.3 45.6 41.9 38.9 35.6
27 49.6 47 43.2 40.1 36.7
28 51.0 48.3 44.5 41.3 37.9
29 52.2 49.6 45.7 42.6 39.1
30 53.7 50.9 47 43.8 40.3
40 66.8 63.7 59.3 55.8 51.8
50 79.5 76.2 71.4 67.5 63.2
60 92.0 88.4 83.3 78.1 74.4
70 104.2 100.4 95 90.5 85.5
80 116.3 112.3 106.6 101.9 96.6
90 128.3 124.1 118.1 113.1 107.6
 96

100 140.2 135.8 129.6 124.3 118.5

Graded Questions

1. Use the chi-square test with =0.05, to test whether the data shown below are
uniformly.
Distributed.

0.34 0.90 0.25 0.89 0.87 0.44 0.12 0.21 0.46 0.67
0.83 0.76 0.79 0.64 0.70 0.81 0.94 0.74 0.22 0.74
0.96 0.99 0.77 0.67 0.56 0.41 0.52 0.73 0.99 0.02
0.47 0.30 0.17 0.82 0.56 0.05 0.45 0.31 0.78 0.05
0.79 0.71 0.23 0.19 0.82 0.93 0.65 0.37 0.39 0.42

2 Use the linear congruential method to generate a sequence of three two-digit random
integers. Let Xo = 27, a = 8, c = 47 and m = 100.

3. Do we encounter a problem in the previous exercise if Xo = O?

4. Use the multiplicative congruential method to generate a sequence of

four three- digit random integers. Let Xo = 117, a = 43 and m = 1000.

5. The sequence of numbers 0.54, 0.73, 0.98, 0.11 and 0.68 has been generated. Use
the kolmogorov-
Smirnov test with  = 0.05 to determine if the hypothesis that the numbers are
uniformly distributed
on the interval [0, 1] can be rejected.

6. Consider the first 30 two digit values in Example 1. Based on runs up

and runs down, determine whether the hypothesis of independence
can be rejected, where  = 0.05.

7. Consider the first 30 two digit values in Example 1. Determine whether

there is an excessive number of runs above or below the mean. Use 
= 0.05.
 97

8. Consider the first 30 two digit values in Example 1. Test whether the
2nd, 9th, 16th, … numbers in the sequence are autocoerrelated, where 
=0.05.

9. A sequence of 1000 four-digit numbers has been generated and an

analysis indicates the following combinations frequencies. Four
different digits, 565; one pair 392; two pair 17; 3 like digit, 24; 4 like
digit, 2. Based on the Poker test, Test whether these numbers are
independent. (Use  = 0.05)

10. Use the mixed congruential method to generate a sequence of

three two-digit random numbers with Xo = 37, a = 7, c = 29, and m =
100.

11. Use the mixed congruential method to generate a sequence of three

two-digit random integers between 0 and 24 with Xo = 13, a = 9, c= 35.

12. Consider the multiplicative congruential generator under the

following circumstances:
(a) a = 11, m = 16, Xo = 7.
(b) A = 11, m= 16, Xo = 8.
(c) A = 7, m = 16, Xo = 7.
(d) A = 7, m = 16, Xo = 8.
Generate enough values in each case to complete a cycle. What
inference can be drawn? Is maximum period achieved?

13. For 16-bit computers, combine three multiplicative generators with m1

= 32363, a1 = 157, m2 = 31727, a2 = 146, m3 = 31657 and a3 = 142.
Generate 5 random numbers with the combined generator using intial
seeds Xi,o = 100, 300, 500 for the individual generators i = 1,2,3.

14. Test the following sequence of numbers for uniformity and

independence using produres you learned in this chapter: 0.574,
0.928, 0.515, 0.055, 0.507, 0.351, 0.262, 0.797, 0.788, 0.442, 0.097,
0.798, 0.227, 0.127, 0.474, 0.825, 0.007, 0.182, 0.929, 0.852.
 98

15. Consider the data in example 1. Can the hypothesis that the numbers
are. Independent be rejected no the basis of the length of the runs up
and down when  = 0.05.

16. State the various test for random numbers and explain briefly any two
of them.

17. Mention the properties of random numbers, and give the methods of
generating pseudo random numbers.

Summary :-
In the chapter the desoubtion of the generation of random nos. are given and the
generated nos. are uniform & independent.

Of the many types of random no. generaters available, the linear

congruential method is the most widely used. Many types of statistical
tests are used in testing random number generators, same of these tests
are for uniformity & others for testing independence.


 99

8
Random Variate Generation

Here we deal with procedures for sampling from a variety of continuous

and discrete distributions, it is assumed that a distribution has been
completely specified.

Inverse Transform Techniques:

The Inverse transform technique can be used to sample from a variety of distributions.
The technique is described in the following steps.

Step 1: Compute the cdf of the desired r, v, x.

Step 2 : Set F(x) = R on the rance of , X, where R represents random

number which is uniformly
distributed on (0,1).

Step 3 : Solve the equation F(x) = R

i.e. X = F-1 (R)

Step 4 : Generate uniform random numbers R1, R2, … and compute the
desired random variates by
Xi = F-1 (Ri)

(a) Exponential distribution :

The exponential sdistribution has pdf

F(x) =  e-x, x  0
=0 otherwise
the parameter  can be interpreted as the mean number of occurances
per unit time. We with to generate values X1, X2,…which have an
exponential distribution.
 100

Step 1 : F(x) = P (X  x) = 1 - ex, x  0

Step 2 : set, 1 – e-x = R, x  0

Step 3 : 1 – e -x = R.

e-x = 1-R

 -x = loge (1-R)

x=
1
loge (1-R)


Step 4 : Generate uniform random numbers R1, R2, … and compute the
desired random variates by Xi = F –1 (Ri)

1
Here, Xi = loge (1-Ri), i = 1,2,3,….


(b) Uniform distribution:

Consider a random variable X that is uniformly distributed on the

interval [a,b]. The pdf of X is given by

f(x) = 1/b-a, a  x  b

= 0, otherwise

Step 1 : The cdf is given by

0, x<a

F (x) = x – a/ b – a , a  x  b

= 1, x>b

Step 2 : set F(x) = x-a / b-a = R.

 101

Step 3 : Solving for X we get, x = a + (b-a) R

(c) Weibull distribution :

the weibull distribution may be used as a model for time to failure

machinics or electronic components. If v = 0, its pdf is given by

  1 ( x / a ) 
f (x) = x e , x0


= 0, otherwise
10. Data have been collected on service times at a drive-in-bank window at the Shady
Lane National Bark. This data are summarized into intervals as follows:

Interval Frequency
(Seconds)
15-30 10
30-45 20
45-60 25
60-90 35
90-120 30
120-180 20
180-300 10
Set up a table for generating service times by the table-look up method and generate
five values of service time using four-digit random numbers.

11. For a preliminary version of a simulation model, the number of pallets,

X to be loaded onto a truch at a loading dock was assumed to be
uniformly distributed (discrete) between 8 and 24. Devise a method
for generating X, assuming that the load of successive trucks are
independent. Generate load for 4 successive trucks by using four digit
random numbers.

12. The weekly demand, X for a slow moving item has been found to be
well approximated by a geometric distribution on the range { 0,1,2,…}
 102

with mean weekly demand of 2.5 items. Generae 10 values of X,

demand per week, using random numbers from Table.

13. Suppose that the demand has been found to have a Poisson
distribution with mean 0.5 items per week. Generate 3 values of X,
Demand per week, using random numbers from table. Use four digit
random numbers.

14. Lead times have been found to be exponentially distributed with mean
3.7 days. Generate five random lead times from this distribution.

15. Develop a technique for generating a binomial random variable, X,

using the convolution technique.

16. The beta (4,3) distribution has pdf (X) = 60x3 (1-x)2, 0  x  1
= 0, otherwise

Generate two variate values from the above distribution using

acceptance-rejection technique. Use three digit random numbers.
Random Digits

94737 08225 35614 24826 88319 05595 58701 57365 74759

87259 85982 13296 89326 74863 99986 68558 06391 50248
63856 14016 18527 11634 96903 52146 53496 51730 03500
66612 54714 46783 61934 30258 61674 07471 67566 31635
30712 58582 05704 23172 86689 94834 99057 55832 21012
69607 24145 43886 86477 05317 30445 33456 34029 09603
37792 27282 94107 41967 21425 04743 42822 28111 09757
01488 56680 73847 64930 11108 44834 45390 86043 23973
66248 97697 38244 50918 55441 51217 54786 04940 50807
51453 03462 61157 65366 61130 26204 15016 85665 97714
92168 82530 19271 86999 96499 12765 20926 25282 39119
36463 07331 54590 00546 03337 41583 46439 40173 46455
47097 78780 04210 87084 44484 75377 57753 41415 09890
80400 45972 44111 99708 45935 03694 81421 60170 58457
94554 13863 88239 91624 00022 40471 78462 96265 55360
31567 53597 08490 73544 72573 30961 12282 97033 13676
 103

07821 24759 47266 21747 72496 77755 50391 59554 31177

09056 10709 69314 11449 40531 02917 95878 71587 60906
19922 37025 80731 26179 16039 01518 82697 73227 13160
29923 02570 80164 36108 73689 26342 35712 49137 13482
29602 29464 99219 20308 82109 03898 82072 85199 13103
94135 94661 87724 88187 62191 70607 63099 40494 49069
87926 34092 34334 55064 43152 01610 03126 47312 59578
85039 19212 59160 83537 54414 19856 90527 21756 64783
66070 38480 74636 45095 86576 79337 39578 40851 53503
78166 82521 79261 12570 10930 47564 77869 16480 43972
94672 07912 26153 10531 12715 63142 88937 94466 31388
56406 70023 27734 22254 27685 67518 63966 33203 70803
67726 57805 94264 77009 08682 18784 47554 59869 66320
07516 45979 76735 46509 17696 67177 92600 55572 17245
43070 22671 00152 81326 89428 16368 57659 79424 57604
36917 60370 80812 87225 02850 47118 23790 55043 75117
03919 82922 02312 31106 44335 05573 17470 25900 91080
46724 22558 64303 78804 05762 70650 56117 06707 90035
16108 61281 86823 20286 14025 24909 38391 12183 89393
74541 75808 89669 87680 72758 60851 55292 95663 88326
82919 31285 01850 72550 42980 57518 01169 01786 98145
31388 26809 77258 99360 92362 21979 41319 75739 98082
17190 75522 15687 07161 99745 48767 03121 20046 28013
00466 88068 68631 98745 97810 35886 14497 90230 69264

Direct transformation for the Normal & Log normal distributions-

The standard normal cdf is given by

Ø (x) = ∫ x∞ 1/ √2 Ï e-t2/2 dt, -∞<x<∞

An initiatively appealing direct transformation that produces an

independent pair of standard normal varieties with mean zero & variance
1.
Transformation is required to transform standard normal
variety into a normal variety with mean x & variance 62.
 104

Acceptance – Rejection technique

Suppose that an analyst needed to devise a method for generating

random variables X uniformly distributed between ¼ & 1.
The steps used are-
a) Step 1- Generate a random no. R
Step 2- a) of R > ¼, accept X = R, then go to step 3.

b) Of R < ¼, reject R & return to step 1

Step 3- Of another uniform random variable on [1/4, 1] is needed repeat

the procedure beginning at step 1, or not stop.

Summary:-
The basic principles of random variable generation using the inverse
transform technique, the convolution method & acceptance rejection
techniques have been introduced.


 105

9
Input Modeling

To carry out a simulation using random inputs, we have to specify their

probability distributions e.g. in the simulation of a queucing system, typical input data are
the distribution of time between arrivals and service times. Given that random variables
follow particular distributions, the simulation proceeds through time by generating random
from these distributions.

Almost all real-world system contains one or more sources of randomness. However,
determining appropriate distributions for input data is a major task from the standpoint of
time and resource requirements. Faulty models of the inputs will lead to outputs whose
interpretation may give rise to misleading conclusions.

There are four in the development of a useful model of input data.

1. Collect data from real system of interest. This requires substantial time and
resources. In some situations it may not be possible to collect data. When data
are not available, expert opinion and knowledge of the process must be used to
make educated guesses.
2. Identify a probability distribution to represent the input process. Here, a
frequency distribution or histogram is developed. Based on the frequency
distribution and structural knowledge of the process, a family of distribution is
chosen.
3. Choose parameters that determine a specific instance of the distribution family.
When data are available, these parameters may be estimated from the data.
4. Evaluate the chosen distribution and the associated parameters for goodness-of-
fit. It the chosen distribution is not a good approximation of the data, then the
analyst chosen a different family of distribution, and repeats the procedure. If
several iterations of this procedure fail to yield a fit between an assumed
distributional form and the collected data the empirical form of the distribution
may be used.

Data Collection:
 106

Data collection is one of the biggest tasks in solying a real problem. It is one of the most
important and difficult problems in simulation. Even when data are available, they may
not be recorded in a form that is directly useful for simulation input modeling. Even if the
model structure is valid, if the input data are inaccurately collected, inappropriately
analysed or not representative of the environment, the simulation output data will be
misleading when used for policy making.

The following points may help in data collection:

1. Plan the data collection properly. Begin by a practice or preobserving session.

Try to collect data while preobserving. Devise forms for this purpose. It is very
likely that these forms will have to be modified several times before the actual
data collection begins. Watch for unusal circumstances and consider how they
will be handled.
2. Try to analyze the data as they are being collected. Determine if the data being
collected are adequate to provide the distributions needed as input to the
simulation. Determine if any data being collected are useless to the simulation.
There is no need to collect superfluous data.
3. Try to combine homogeneous data sets. Check data for homogeneity in
successive time periods and during the same time period on successive days.
4. Be aware of the possibility of data censoring, in which a quantity of interest is not
observed in its entirety. This problem most often occurs when the analyst is
interested in the time required to complete some process, but the process begins
prior to, or finishes after the completion of, the observation period.
5. To determine whether there is a relationship between two variables, build a
scatter diagram. Sometimes an eyeball scan of the scatter diagram will indicate if
there is a relationship between two variables of interest.
6. Consider the possibility that a sequence of observations which appear to be
independent may possess autocorrelation. Autocorrelation may exist in
successive time periods or for successive customers.
7. Keep in mind the difference between input data and output or performance data,
and be sure to collect input data. Input data typically represent the uncertain
quantities that are largely beyond the control of the system and will not be altered
by changes made to improve the system. Output data on the other hand
represent the performance of the system when subjected to the inputs,
performances that we may be trying to improve.

Identifying the Distribution with data:

 107

There are several methods for selecting families of input distributions when data are
available. The specific distributions within a family are specified by estimating its
parameters.

(1) Histograms:

A frequency distributions or histogram is useful in identifying the shape of a distribution. A

histogram is constructed as follow:

1. Divide the range of the data into intervals (intervals are usually of equal width;
however, unequal widths may be used if the heights of the frequencies are adjusted)
2. Label the horizontal axis to conform to the intervals selected.
3. Determine the frequency of occurrences within each interval.
4. Label the vertical axis so that the total occurrences can be plotted for each interval.
5. Plot the frequencies on the vertical axis.

The number of class intervals depends on the number of observations and the
amount of scatter or dispersion in the data. Choosing the number of class intervals
approximately equal to the square root of the sample size often works well in practice. If
the intervals are too wide, the histogram will be coarse, or blocky, and its shape and
other details will not show well. If the intervals are too narrow, the histogram will be
ragged and will not smooth the data.

The histogram for continuous data corresponds to the probability density function of a
theoretical distribution, if continuous, a line through the center point of each class interval
frequency should result in a shape like that of a pdf.

(2) Selecting the family of distributions:

A family of distribution is selected on the basis of what arise in the context being
investigated along with the shape of the histogram.
Binomial: Models the number of successes in n trials, when the trials are independent
with common success probability, p.
Poisson: Models the number of independent events that occur in a fixed amount of time
or space.
Normal: Models the distribution of a process that can be thought of as the sum of a
number of component processes.
Exponential: Models the time between independent events, or a process time which is
memoryless.
Gamma: An extremely flexible distribution used model nonnegative random variables.
 108

Beta: An extremely flexible distribution used to model bounded random variables.

Erlang: Models processes that can be viewed as the sum of several exponentially
distributed processes.
Weibull: Models the time to failure for components. The exponential is a special case of
the weibull.
Discrete or continuous Uniform: Models complete uncertainty; since all the outcomes
are equally likely.
Triangular: Model a process when only the minimum, most-likely, and maximum values
of the distribution are known.
Empirical: Resamples from the actual data collected; often used when no theoretical
distribution seems appropriate.
Do not ignore physical characteristics of the process when selecting distributions. Is the
process naturally discrete or continuous valued? Is it bounded or is there no natural
bound? This knowledge, which does not depend on data, can help narrow the family of
distributions from which to choose.

Graded Questions

1. The following data are randomly generated from gamma distributions:

1.691 1.437 8.221 5.976
1.116 4.435 2.345 1.782
3.810 4.589 5.313 10.90
2.649 2.432 1.581 2.432
1.843 2.466 2.833 2.361
Determine the maximum-likelihood estimators B and A.

2. The highway between Atlanta, Georgia and Athens, Georgia, has a high
incidence of accidents along its 100 km. Public safety officers say that the
occurrence of accident along the highway is randomly (uniformly) distributed, but
the news media says otherwise. [N-04] The Georgia department of Public Safety
published records for the month of September.

These records indicated the point at which 30 accidents occurred, as follows:

88.3 40.7 36.3 27.3 36.3

91.7 67.3 7.0 45.2 23.3
98.8 90.1 17.2 23.7 97.4
32.4 87.8 69.8 62.6 99.7
20.3 73.1 21.6 6.0 45.3
76.6 73.2 27.3 87.6 87.2
 109

Use the Kolmogorov- Smirnov test to determine whether the distribution is

uniformly distributed given D0.05, 30 = 0.24

3. Records pertaining to the monthly number of job-related injuries at an

underground coal mine were being studied by a federal agency. The values for
the past 100 months were as follows.

Injuries per Month Frequency of Occurrence

0 35
1 40
2 13
3 6
4 4
5 1
6 1

(a) Apply the chi-square test to these data to test the hypothesis that the underlying
distributions are Poisson. Use a level of significance of a = 0.05.

(b) Apply the chi-square test of these data to test the hypothesis that the distribution is
Poisson with mean 1.0. Again let a = 0.05.

4. The time required for 50 different employees to conpute and record the number of
hours worked during the week was measured with the following results in minutes.

Employee Time (Minutes) Employee Time (Minutes)

1 1.88 26 0.04
2 0.54 27 1.49
3 1.90 28 0.66
4 0.15 29 2.03
5 0.02 30 1.00
6 2.81 31 0.39
7 1.50 32 0.34
8 0.53 33 0.01
9 2.62 34 0.10
10 2.67 35 1.10
11 3.53 36 0.24
12 0.53 37 0.26
13 1.80 38 0.45
14 0.79 39 0.17
 110

15 0.21 40 4.29
16 0.80 41 0.80
17 0.26 42 5.50
18 0.63 43 4.91
19 0.36 44 0.35
20 2.03 45 0.36
21 1.42 46 0.90
22 1.28 47 1.03
23 0.82 48 1.73
24 2.16 49 0.38
25 0.05 50 0.48

Use the chi-square test to test the hypothesis that these service times are exponentially
distributed. Let the number of class intervals be k = 6. Use a level of significance of a =
0.05.

5. 200 electric High bulbs were tested and the average lifetime of the bulbs was found to
be 25 hrs. Using the summary given below, test the hypothesis that the lifetime is
exponentially.

Lifetime in hrs: 0-20 20-40 40-60 60-80 80-

No. of bulbs : 104 56 24 12 4

6. Explain time series input models with suitable examples. [N-04]

7. The following data were available for the past 10 years on demand and lead-time.
[N-04]
Lead time: 6.5 4.3 6.9 6.9 6.9 6.9 5.8 7.3 4.5 6.3
Demand: 103 83 116 97 112 104 106 109 92 96

Estimate correlation and covariance.

8. Discuss the steps involved in the development of a model of input data. [M-05]

Parameter Estimation: -

1) Sample Mean & Sample variance-

In a number of instances the sample mean or the sample variance are used to estimates
the parameters of a hypothesized distribution.
The data are discrete & have been grouped in a frequency distribution.
Of the observations in a sample of size n are X1, X2….Xn the sample mean (X)
 111

 Xi
n

By X  i 1

n
2
& the sample variance S is defined by

X  nX
n
2 2
i
S2  i 1

n 1
And if the data are direct & grouped in a frequency distribution then sample mean can be
computed by

 fiXj
k

X 
j 1

n
& sample variance by

 fiXj  n X
k
2

S2 
j 1

n 1
Where k is the no. of distinct values of X &
fi is the observed frequency of the value Xj of X?

11. Suggested Estimators for distributions often used in simulation-

Distribution Parameter Suggested Estimator


Poisson  x
Exponential   1 X

Gamma  ,  1 X

Normal , 2 x

  S2
 112


Lognormal , 2 x

  S2

Weibull ,  0  X S
Taking  0

Goodness of Fit Tests-

For the goodness of fit tests, kolmogorov-& mirnov test and the chi-square test
were introduced these two tests are applied in this section to hypotheses about
distributional forms of input data.

(i) Chi square test -

One procedure for testing the hypotheses that a random sample of size n of the random
variable X follows a specific distributional form is the chi square goodness of fit test.
The test is valid for large sample sizes for both discrete & continuous distributional
assumptions.

The test statistic is given by

(Oi  Ei ) 2
X0
K
2

i 1 Ei

Oi is observed frequency in the ith class Ei is the expected frequency in the class interval.
Ei = npi
Pi is the theoretical hypothesized probability associated with the ith class interval.

(ii) Kolmogorov-Smirnov Goodness of fit test –

The chi square goodness of fit test cans a common date the estimation of parameters
from the data with resulted decreases in the degree of random.
The k-s test is particularly useful when sample sizes are small and when no parameters
have been estimated from the data.
K-s test is also applicable to the explain distribution.

Covariance & Correlation

Let X1 & X2 be two random variables, & let

 113

i  E ( Xi ) &  i  var( Xi ) be the mean & variance of Xi respectively.

2

The covariance and correlation are measures of the linear dependence between X1 &

 ( X 1   )   ( x2   ) + E
X2.

1 2
Where E is a random variable with mean O that is independent of X2.

 ( X 1   )   ( x2   )
.

1 2

Then the model is perfect.

And if X 1 & X 2 are independent then B=0.

[( X 1   )( x2   )]
The covariance between X1 & X2 is defined to be
C or ( X 1 , X 2 ) = E
1 2

=E ( X 1 X 2 )  1  2

A value of cove ( X 1 X 2 ) = 0 implies B=0 then our model is dependent

The covariance can take any value between. X and x. The correlation standardizes the
covariance to bet between -1 & 1

 P  corr ( X 1 X 2 ) 
cov( X 1 X 2 )
 1 2

Summary –

Input data collection and analysis require major time and resources commitments in a
discrete event simulation project.
In this chapter, four steps in the development of models of input data is discussed

1) Collecting the raw data

2) Identifying the underlying statistical distribution.
3) Estimating the parameters.
4) Testing for goodness of fit.
 114



10
Verification and Validation of Simulation Model

One the most difficult problems facing a simulation analyst is that of trying to determine
whether a simulation model is an accurate representation of the actual being studied, i.e.
whether the model is valid.

Verification is concerned with determining whether the conceptual simulation model

(model assumptions) has been correctly translated into a computer program.
Validation is the process of determining whether a simulation model is an accurate
representation of the system, for the particular objectives of the study.

Model Building, Verification and Validation

The first step in model building consists of observing the real system and the interactions
among its various components and collecting data. Observation alone rarely gives
sufficient understanding of system behavior. Persons familiar with the system should be
interviewed to get information about the system. Model building is a continuous process.
The second step in model building is the construction of a conceptual model. This
includes assumptions on the components and the structure of the system and the values
of input parameters.

The third step is the translation of the operational model into a computer recognizable
form that is a computerized model.

The model builder will repeat these three steps many times, till an acceptable model is
achieved.

Verification of Simulation Model

The purpose of model verification is to assure that the conceptual model is reflected
accurately in the computerized representation.
 115

Some observations to help the process of verification:

1. Get the computer representation checked by an independent and knowledgeable

person.

2. Make a detailed flow diagram.

3. Closely examine the model output for reasonableness under a variety of setting
of the input parameters. Computer model should print a variety of output
statistics along with input parameters.

4. Prepare a detailed documentation. Give precise definition of every variable used.

5. An interactive run controller (IRC) or debugger is an important for simulation

model building. One of the most powerful techniques to debug a discrete-event
simulation is a trace in a trace; the contents of the event list, the state variables,
certain statistical counters etc. are displayed just after each event occurs. These
values value are compared with hand calculations to see if the program is
operating as intended.

Calibration and Validation of Models

Validation is the overall process of comparing the model and its behavior to the real
system and its behavior. Calibration is the iterative process of comparing the model to the
real system. Making adjustment or changes to the model, comparing the revised model to
reality and so on.

The comparison of the model to reality is carried out by subjective and objective tests. A
subjective test involves talking to people, who are knowledgeable about the system,
making models and forming the judgment. Objective tests involve one or more statistical
tests to compare some model output with the assumptions in the model.

Naylor and Finger formulated a three step approach to the validation process.

1. Build a model that has high face validity.

2. Validate model assumptions.

3. Compare the model input-output transformations to corresponding input-output

transformations for the real system.

Face Validity
 116

The first goal of the simulations modeler is to construct a model that appears reasonable
on its face to model users are others who are knowledgeable about the real system being
simulated. The potential users of a model should be involved in model construction from
its conceptualization to its implementation to ensure that a high degree of realism in built
into the model through reasonable assumptions regarding system structure, and reliable
data. Potential users and knowledgeable persons can also evaluate model output for
reasonableness and can aid in identifying model deficiencies. Thus the users can be
involved in the calibration process as the model is iteratively improved, based on the
insights gained from the initial model deficiencies. Another advantage of user
involvement is the increase in the model’s perceived validity, or credibility, without which
a manager would not be willing to trust simulation results as a basis for decision making.

Sensitivity analysis can also be used to check a model’s face validity. The model user is
asked if the model behaves in the expected way when one or more input variables are
changed.

Validation of Model Assumptions

Model assumptions fall into two general classes. Structural assumptions and data
assumptions. Structural assumptions involve questions of how the system operates and
usually involve simplifications and obstructions of reality.

Data assumptions should be based on the collection of reliable data and correct
statistical analysis of the data.

When combining two or more data sets collected at different times, data reliability can be
further enhanced by objective statistical tests for homogeneity of data.

Additional tests may be required to test for correlation in the data. As soon as the analyst
is assured of dealing with a random sample (i.e. correction is not present), the statistical
analysis can begin.

The procedures for analyzing input data from a random sample were discussed in detail
in chapter 9. Whether by hand, or using computer software for the purpose, the analysis
consists of three steps.

1. Identifying the appropriate probability distribution.

2. Estimating the parameters of the hypothesized distribution.

3. Validating the assumed statistical model by a goodness-f-fit test, such as the chi-
square or Kolmogorov-Smirnov test, and by graphical methods.

The use of goodness-of-fit tests is an important part of the validation of the model
assumptions.
 117

Validating Input-Output Transformations

The ultimate, test of a model, and in fact the only objective test of the model as a whole,
is its ability to predict the future behavior of the real system when the model input data
match the real inputs and when a policy implemented in the model is implemented at
some point in the system. Furthermore, if the level of some input variables (e.g. the
arrival rate of customers to a service facility) were to increase of decrease, the model
should accurately) predict what would happen in the real system under similar
circumstances. In other words, the structure of the model should be accurate enough for
the model to make good predictions, not just for one input data set, but for the range of
input data sets which are of interest.

In this phase of the validation process, the model is viewed as an input-output

transformation. That is the model accepts values of the input parameters and transforms
these inputs into output measures of performances. It is this correspondence that is being
validated. Instead of validating the model input-output transformations by predicting the
future, the modeler may use past historical data which have been reserved for validation
purposes only; that is, if one data set has been used to develop and calibrate the model,
it is recommended that a separate data set be used as the final validation test. Thus,
accurate “prediction of the past” may replace prediction of the future for the purpose of
validating the model.

A necessary condition for the validation of input-output transformation is that some

version of the system under study exists, so that system data under at least one set of
input conditions can be collected to compare to model predictions. If the system is in the
planning stages and no system operating data can be collected complete input-output is
not possible.

Input-Output Validation Using Historical Input Data

An alternative to generating input data is to use the actual historical record, {An, Sin, n=1,
2….} to drive the simulation model and then to compare model output to system data.

When using this technique the modeler hopes that the simulation will duplicate as closely
as possible the important events that occurred in the real system.

To conduct a validation test using historical input data, it is important that all the input
data (An, Sn,) and all the system response data, such as average delay (Z2), be
collected during the same time period. Otherwise, the comparison of model responses to
system responses, such as the comparison of average delay in the model (Y2) to that in
the system (Z2), could be misleading. The responses (Y2 and Z2) depend on the inputs
(An and Sn) as well as on the structure of the system, or model. Implementation of this
technique could be difficult for a large system because of the need for simultaneous data
collection of all input variables and those response variables of primary interest. In some
 118

systems, electronic counters and devices are used to ease the data-collection task by
automatically recording certain types of data.

Input-Output Validation Using a Turing Test

In addition to statistical tests, or when no statistical test is readily applicable persons
knowledgeable about system behavior can be used to compare model output to system
output. For example, suppose that five reports of system performances over five different
days are prepared, and simulations output data are used produce five additional reports.
The 10 reports should all be in exactly the same format and should contain information of
the randomly shuffled and given to the engineer, who is asked to decide which reports,
the model builder questions the engineer and uses the information gained to improve the
model. If the engineer cannot distinguish between fake and real reports with any
consistency, the modeler will conclude that this test provides no evidence of model
inadequacy. This type of validation test is called a Turing Test.
Graded Questions
1. A simulations model of a job shop was developed to investigate different
scheduling rules. To validate the model, the currently used scheduling rule was
incorporated into the model and the resulting output compared to observed
system behavior. By searching the previous year’s computerized records it was
estimated that the average number of jobs in the shop was 22.5 on a given day.
Seven independent replications of the model were run each pf 30 days duration,
with the following results for average number of jobs in the shop.
18.9 22.0 19.4 22.1 19.8 21.9 20.2

Develop and conduct a statistical test to determine if model output is consistent with
system behavior. User a level of significance of a=0.05.

2. System data for the job of exercise 1 revealed that the average time spent by a
job in the shop was approximately 4 working days. The model made the following
predictions on seven independent replications, for average time spent in the
shop.

3.70 4.21 4.35 4.13 3.83 4.32 4.05

Is model output consistent with system behavior? Conduct a statistical test using a level
of significance a=0.01. If it is important to detect a difference of 0.5 day, what sample size
is needed?

3. For the job of exercise 1, four sets of input data were collected over four different
10 day periods, together with the average number of jobs in the shop (Zi) for
each period. The input data were used to drive the simulation model for four runs
 119

of 10 days each and model predictions of average number of jobs in the shop
(Yi) were collected, with these results:

I 1 2 3 4
Zi 21.7 19.2 22.8 19.4
Yi 24.6 21.1 19.7 24.9

Conduct a statistical test to check the consistency of system output and model output.
Use a level of significance of a = 0.05.

4. Explain in detail the 3-step approach of Naylor and finger in the validation
process.

5. How would you validate simulation models?

Percentage Points for the Students Distribution with V Degrees of Freedom

0 l
 120

V t0.005 t0.01 t0.025 t0.05 t0.10

 121

1 63.66 31.82 12.71 6.31 3.08

2 9.92 6.92 4.30 2.92 1.89
3 5.84 4.54 3.18 2.35 1.64
4 4.60 3.75 2.78 2.13 1.53
5 4.03 3.36 2.57 2.02 1.48
6 3.71 3.14 2.45 1.94 1.44
7 3.50 3.00 2.36 1.90 1.42
8 3.36 2.90 2.31 1.86 1.40
9 3.25 2.82 2.26 1.83 1.38
10 3.17 2.76 2.23 1.81 1.37
11 3.11 2.72 2.20 1.80 1.36
12 3.06 2.68 2.18 1.78 1.36
13 3.01 2.65 2.16 1.77 1.35
14 2.98 2.62 2.14 1.76 1.34
15 2.95 2.60 2.13 1.75 1.34
16 2.92 2.58 2.12 1.75 1.34
17 2.90 2.57 2.11 1.74 1.33
18 2.88 2.55 2.10 1.73 1.33
19 2.86 2.54 2.09 1.73 1.33
20 2.84 2.53 2.09 1.72 1.32
21 2.83 2.52 2.08 1.72 1.32
22 2.82 2.51 2.07 1.72 1.32
23 2.81 2.50 2.07 1.71 1.32
24 2.80 2.49 2.06 1.71 1.32
25 2.79 2.48 2.06 1.71 1.31
26 2.78 2.48 2.06 1.71 1.32
27 2.77 2.47 2.05 1.70 1.31
28 2.76 2.47 2.05 1.70 1.31
29 2.76 2.46 2.04 1.70 1.31
30 2.75 2.46 2.04 1.70 1.31
40 2.70 2.42 2.02 1.68 1.30
60 2.66 2.39 2.00 1.67 1.30
120 2.62 2.36 1.98 1.66 1.29
00 2.58 2.33 1.96 1.645 1.28

Summary –
 122

Validation of simulation models is great importance. Decisions are made on the basis of
simulations results thus the accuracy of these results should be subject to question and
investigation.

The same possible validation technique

1. Develop models with high face validity by consulting persons knowledgeable

about system behavior on model structure, model input and model output.

2. Conduct simple statistical tests of input data

3. Conduct a trying tests

4. Compare model o/p to system o/p by means of statistical tests.


 123

11
Output Analysis for a Single Model

Output analysis is the examination of data generated by a simulation. Its purpose is to

predict the performance of a system or to compare the performance of two or more
system designs. Usually random number generators are used to produce the values of
the input variables. Therefore, the output data from simulations exhibits random
variability. That is, two different streams of random numbers will produce two sets of
output data which will differ.

If the performance of the system is measured by O, the result of a set of simulations

experiments will be an estimator O of O. The precision of the estimator O can be
measured by the variance or standard error or O. The purpose of the statistical analysis
is to estimate is variance or to determine the number of observations required to achieve
a desired precision.

A typical output variable, Y, should be treated as a random variable with an unknown

distribution. Let Y1, Y2, Yn be an output from a single simulation run e.g. Yi may be
production in the Itch hour, inventory cost of Itch week etc. The Yi’s are random variables
that will, in general, be neither independent nor identically distributed. Thus, the
sequence of random variables Y1, Y2….Yn may be auto correlated. Auto correlation is a
measure of a lack of statistical independence. This means that the classical methods
(tests) of statistics which assume independence will have to be properly modified.

The specification of initial condition may influence the output data. That is the initial
conditions may influence the values Y1, Y2……..Yn.

For the purpose of statistical analysis, the effect of the initial conditions is that the output
observations may not be identically distributed, and the initial observations may not be
representative of steady state behavior of the system. An incorrect choice of initial
condition may lead to an improper estimate of the steady performance of the simulations
model.

Types of Simulation with respect to output Analysis

Simulations may be either terminating or none terminating, depending on whether there

is an obvious way for determining run length.
 124

A terminating simulation is one that runs for some duration of time. Te, where E is
specified event or set or events which stops the simulations. Such a simulations system
“Opens” at time zero under specified initial conditions and “Closes” at the stopping time
Te, since the initial conditions for a terminating simulation generally affect the desired
measures of performance. These conditions should be representative of those for the
actual system, e.g. simulations of a Bank from 9.00 a.m. (time 0) to 3.00 p.m. (time
Te=360 minutes)

A non terminating simulations is one for which there is no natural event E to specify the
length of a run. E.g. continuous production systems, telephone systems, internet, OPD of
a Hospital, designing a new system.

A Simulation of non terminating system starts at simulation time zero under initial
conditions defined by the analyst and runs for some analyst-specified period of time Te.
The analyst wants to study stead-state or long-run properties of the system-that is,
properties which are not influenced by the initial conditions of the model at time zero. A
steady state simulation is a simulation whose objective is to study long-run, or steady-
state, behavior or a non terminating system.

Whether, a simulation is considered to be terminating or not depending on the objective

of the simulation and the nature of the system.

Stochastic Nature of Output data

Consider one run of a simulation model over a period of time. [è, T]. Since some of the
model inputs are random variables, the model output variables are also random
variables. This indicates stochastic (Probabilistic) nature of output variables.

Measures of performance and their Estimation

Consider the estimation of a performance parameter è (or è), of a simulated system. Let
Y1, Y2……..Yn be the output data which is called discrete-time data. Suppose discrete-
time data are used to estimate è. Let {Y(t), è < Te} be the output data which is called
continuous-time data continuous-time are used to estimate è. e.g. Yi may be total time
spent in system by customer i. Y (t) may be the number of customers in the system at
time t.

0 is an arithmetic mean and o is called time-weighted mean.

Point Estimation

 Yi


1 n
The point estimator of O based on Y1, Y2……..Yn is defined by
n m
 125

The point estimator __ is said to be unbiased for __ if E(_) = __. E(_) - __ is called the
bias in __.

The point estimator  is given by

TE 0
 T


i 1
Y (t )dt

Interval estimation

   
Let  be a point estimator based on Y1, Y2, …..Yn. Let ó2(  ) = v(  ). Let ó2 (  ) be an

estimator of ó (  ).
2


  
If ó (  ) is approximately unbiased, then the statistic t
2

is approximately t-

 ( )
distributed with degrees of freedom f(say). An approximate 100(1- )% confidence
  
interval for  is given by   t a 2 , r. ( ), t a 2 , r is such that P[t  t a 2 , r]   2 .

Output Analysis for Terminating Simulations

The Simulation is repeated R times, each run using a different random-number stream
and independently chosen initial conditions.

observation with replication r, for I =1, I =2, …. Nr, and r = 1, 2, …. R.

th
Let Yn be the n

Y

r 
n1
, r  1,2,....R.
1
The sample mean for replication r is given by ri
nr m

 
The R sample means  1 ,....... R are statistically independent and identically distributed

and are unbiased simulators of .

 126

Confidence – Interval estimation for a fixed number of replications.

Suppose that R independent replications are made.


The overall point estimate,  is given by,

 r
 1 R 
r 
R r 1

 r )2
2  R  
 ( r )    
1
(
R ( R  1) r 1
and

A 100 (1-) % confidence interval is calculated with f = R-1.

When the output data are of the form ** independent replication, we have,

 Y (t )dt, r  1,......, R...

 TE

r 
1
r
TE 0

 r
 1 R 
r 
R r 1


2  R  
 ( r )    
1
( )2
R ( R  1) r 1
and r

A 100(1-confidence interval is obtained with f = R-1.

Graded Questions

1. Explain point estimation and interval estimation of è (or è).

2. Discuss the output analysis for terminating simulations and confidence interval
estimation for a fixed number of replications.
3. Discuss output analysis for steady state simulations.

Output Analysis for steady state simulations

The steady state measure of performance to be estimated is defined by-

 127

 1 n
  Yi
n   n i 1

With probability 1, where the value of è is independent of the initial conditions.

1. First the initialization bias in steady state simulation is required.

The first method is to initialize the simulation in a state that is more

representative of long seen conditions this method is sometimes called intelligent
initialization.

2. Statistical Background is required.

 con(Y , Y )
2  
 ( )  var( ) 
n n
1
i j
n2 l 1 j 1
2
where con(Yi , Y j )  var(Yi )

3. Replications method for steady state simulations for replication r, define

Yr , (n, d )  Y
n
1
nd
rj
j  d 1

Y1, (n, d )....., YR , (n, d )

 n ,d  E[Yr (n, d )]

The overall point estimator is given by

 Yr ( n , d )
1 R
Yr , ( n, d ) 
R r 1

Sample variance-


1 R
S  2
(Yr  Y .....) 2
R  1 r 1
 128

( Y 2 r  RY 2 ....)
R

1
R  1 r 1

The standard error is given by

s.e.(Y ..) 
S
R

4. Sample size in steady state simulations suppose it is desired to estimate a long

run performance measures è within I.E with confidence 100 (1-<) %

5. Batch means for internal Estimation in steady state simulation

Yj  
1 jm
Y
i  ( j 1) m 1 i  d
m
Sample mean

Y  KY
K
2 2

(Y j  Y )
 
2 2 j


K
S 1 j 1

K K j 1 K  1 K ( K  1)

6. Confidence intervals for Quantizes

R quantile estimates Q1….QR are independent & identically distributed. Their average

 i
 1 R 

R i 1

Confidence Interval-


  t 2, R 1
S
R
 
When S2 is the usual sample variance of  1 ........ R
Summary:
 129

This chapter emphasized the idea that a stochastic disccrete event simulation is a
statistical experiment.

The statistical precision of point estimators can be measured by a standard error

estimate or a confidence interval.

The main point is hat simulation o/p data contain some amount of random variability &
without some assessment of its magnitude; the pf estimate cannot be used with any
degree of reliability.




12
Comparison and Evaluation of Alternative System Design

One of he most important uses of simulation is the comparison of alternative system

designs. Since the observations of the response variables contain random variation,
statistical analysis is needed to determine whether any observed differences are due to
differences in design or merely to the random fluctuation inherent in the models.

Comparison of Two System Designs

Suppose that a simulation analyst desires to compare two possible configurations of a

system. In measure for system I will be denoted by èi (I = 1, 2). If it is a steady state
simulation, it will be assumed that deletion of data, or other appropriate techniques, have
been used to assure that the point estimators are approximately unbiased estimators of
the mean performance measures èi. The goal of the simulation experiment is to obtain
point and interval estimates of the difference in mean performance, namely O1-O2.

When comparing two systems, the simulation analyst must decide on a run length T (i) E
for each model (i=1, 2) and a number of replication RI to be made of each model. From
replication r of system I, the simulation analyst obtains an estimate Yn of the mean
 130

performance measure, èi. Assuming that the estimators Yri are (at least approximately)
unbiased, it follows that

 1  E (Yr1 ), r  1,......R1 ;  2  E (Yr 2 ), r  1,......R2 ;

A two-sided 100 (1-)% c.i. for 1   2 will always be of the form

(Y1,  Y2 )  t a 2 , v s.e. (Y1,  Y2 )

Where Y1 is the sample mean performance measure for system I over all replications.

Y
R1
Yi1 
1
ri
Ri r 1
And v is the degrees of freedom associated with the variance estimator t /2, v is the 100
(1-/2) percentage point of a t-distribution with v degrees of freedom, and S.E. (.)
represents the standard error of the specified point estimator.

If the confidence interval for 1-2 is totally to the left (right) of zero then we may
conclude that
i < 2 (1>2).

If the confidence interval for 1-  contains zero there is no statistical evidence that one
system design is better than the other.

Independent Sampling with Equal Variances

Independent sampling means that different and independent random-number streams will
be used to simulate the two systems. This implies that all the observations of simulated
system 1, namely {Yri, r=1, R1}, are statistically independent of all the observations of
simulated system 2, namely {Yr2, r=1…... R2}. The independence of the replications, the
variance of the sample means Yi is given by

var(Yri )  i
2

var(Yi )   i=1,2
Ri Ri

When using independent sampling Y1 and Y2 are statistically independent; hence,

 131

var(Yi  Y2 )  var(Y1 )  var(Y2 )

 12  22
 
R1 R2

In some cases it is reasonable to assume that the two variances are equal (but unknown
in value) that is  12 =  22

In a steady state simulation, the variance  12 decreases as the run length TE(1)
increases; therefore, it may be possible to adjust the two run length TE(1) and TE( 2 ) , to
achieve at least approximate equality of  12 and  22 ..

If it is reasonable to assume a1 2 = a 2 2 (approximately), a two sample t confidence

interval approach can be used. The point estimate of the mean performances difference
is  i   2 = Yi  Y2

The sample variance for system i is given by

1  R1 2 
   Yri Ri Yi 2 
1 R1
Si   
2
(Y Y )
Ri  1 r 1 Ri  1  r 1 
ri i

Since by assumption  12   22   2 (say), a pooled estimate of  2 is obtained by

( R1  1) S12  ( R2  1) S 22
Sp  which has v  R1  R2  2 degrees of freedom.
2

R1  R2  2

(Yi  Y2 )  S p 
1 1
s.e.
R1 R2

Correlated Sampling or Common Random Numbers

Correlated sampling means that, for each replication, the same random numbers are
used to simulate both systems. Therefore, R1 and R2 must be equal, say R1  R2  R .
Thus for each replication r, the two estimates Yr1 and Yr2 are no longer independent but
rather are correlated.
 132

To compute a 100 (1-  ) % c.i.; with correlated data.

 Dr
1 R1
Let, Dr  Yr1  Yr 2 and D 
R r 1

The sample variance of the differences {Dr} is computed by

1  2
 
R
2

1 R1
( Dr  D ) =  
R  1  r 1 
2 2
SD D R D
R  1 r 1 
r

Which has degrees of freedom v = R – 1

= s.e. (Yi  Y2 ) 
SD
Let s.e. (D )
R

Comparison of Several System Designs

Suppose that a simulation analyst desires to compare K alternative system designs. The
comparison will be made on the basis of some specified performance measure, Oi of
system i for i = 1, 2, K. Many different statistical procedures have been developed which
can be used to analyze simulation data and draw statistically sound inferences
concerning the parameters Oil. These procedures can be classified as being fixed-
sample size procedures, or sequential sampling (or multistage) procedures. In the first
type, a predetermined sample size is used to draw inferences via hypothesis tests or
confidence intervals.

A sequential sampling scheme is one in which more and more data are collected until an
estimator with a prespecified precision is achieved, or until one of several alternative
hypothesis is selected, with the probability of correct selection being larger than a
prespecified value. A two stage 9or multistage) procedure is one in which an initial
sample is used to estimate HW many additional observations are needed to draw
conclusions with a specified precision.

Bonferroni Approach to Multiple Comparisons

Suppose that a total C confidence intervals are computed, and that the ith interval has
confidence coefficient 1- a. Let Si be the statement that the ith confidence interval
contains the parameter (or difference of two parameters) being estimated. This statement
may be true or false for a given set of data, but the procedure leading to the interval is
designed so that statement Si will be true with probability 1- ai. When it is desired to
make statements about several true simultaneously. The Bonferroni inequality states that
 133

 1   j  1   E
C
P (all statements Si are true, I=1, ……, C)
j 1

 E   j
C
Where is called the overall error probability.
j 1

P (one or more statements Si is false I = 1,……… C) < E or equivalently

P (one or more of the confidence intervals does not contain the parameter being
estimated) <  E .

Thus  E provides an upper bound on the probability of a false conclusion. When

conducting an experiment making C comparisons, first select the overall error probability,
say  E = 0.05 or 0.10. The individual aj may be chosen to be equal (  j =  E /C) or
unequal, as desired. The smaller the value of aj, the wider Jth confidence interval will be.

Metamodeling

Suppose that there is a simulation output response variable, Y that is related to k

independent variables, say x1 , x 2 ,.......x k . The dependent variable, Y, is a random
variable, while the independent variables x1 , x 2 ,.......x k . Are called design variables and
are usually subject to control. The true relationship between the variables Y and x is
represented by the simulation model. Our goal is to approximate this relationship by a
simpler mathematical function relationship between Y and x1 , x 2 ,.......x k .say
Y = f ( x1 , x 2 ,.......x k ). However, in most cases, the functional relationship is unknown,
and the analyst must select an appropriate function containing unknown parameters from
a set of data (Y, x). Regression analysis is one method for estimating the parameters.

Optimization via Simulation

Optimization is a key tool used by operations researchers and management scientists,

and there are well developed algorithms for many classes of problems.

In stochastic, discrete-event simulations, the result of any simulation run is a random

variable for notation, let x1 , x 2 ,.......x m .be the m controllable design variable, and let Y
( x1 , x 2 ,.......x m ) be the observed simulation output performance on one run.
 134

What does it mean to optimize Y ( x1 , x 2 ,.......x m ) with respect to x1 , x 2 ,.......x m ? Since

Y is a random variable, we cannot optimize the actual value of Y. The most common
definition of optimization is

Maximize or minimize E (Y( x1 , x 2 ,.......x m ))

In other words, the mathematical expectation, or long run average, of performance is

maximized or minimized. This is the default definition of optimization used in all
commercial packages.

Summary:-

A basic introduction to the comparative evaluation of alternative system designs based

on data collected from simulation runs.

An introduction to metamodels & optimization via simulation was also provided.

Examples:

1. Experimental design models.

2. Output analysis methods.

3. Variance reduction techniques.




13
 135

Simulation of Manufacturing & Material Handling Systems

1. Manufacturing & material handling simulations

(i) Prepare the models of manufacturing systems.

Examples.
a) Physical Layout
b) Labor
c) Equipment
d) Maintenance
e) Workers
f) Product
g) Production control
h) Storage
i) Packing & Shipping

1. Models of Material Handling

Models of material handling system may have to contain some of the following types of
subsystems.

j) Conveyers

ii) Transporters

iii) Storage Systems

2. Some common Material Handling Equipments

There are numerous types of material handling devices common to manufacturing,

warehousing & distribution operations.

Case studies of the Simulation of Manufacturing and material handling systems.

 136

An abstract of some of the papers from the 1994 winter simulation conference process
during will provide some insight into the types of problems that can be addresses by
simulation.

(1) Session: Simulation water Manufacturing

Paper: Modeling & simulation of material handling for semiconductor water

manufacturing.

(2) Session: Simulation in Aerospace manufacturing.

Paper: Modeling Aircraft assembly operations

(3) Session: Control of Manufacturing systems

Paper: Discrete event simulation for shop floor control.

(4) Session: Flexible Manufacturing

Paper: Developing & Analyzing Flexible cell systems using simulation.

Summary:

This chapter introduced some of the ideas & concepts most relevant to manufacturing &
material handling simulation.

The advantages of trace driven simulations with respect to some of the inputs &
the need in some models for accurate modeling of material handling equipment & the
control system.

Case Studies:

Do the case studies of the simulation of manufacturing & material handling systems.


University Exam. Papers – November 2004

Computer Simulation & Modeling

N.B. (1) Question No.1 compulsory.

(2) Attempt any four questions out of remaining six questions.

 137

3. (a) What is system modeling? Give an example and explain the different types of
models.

(b) Define simulation. When is simulation appropriate and when it is not?

4. (a) State the various test for random numbers and explain briefly any two of
them.

(b) What are the characteristics of queuing system and how would you determine
the costs in queuing problems.

3. (a) Name and explain some of the useful statistical models for queuing systems.

(b) The highway between Atlanta, Georgia and Athens, Georgia, has a high incidence of
accident along its 100 km. Public safety officers say that the occurrence of accident along
the highway is randomly (uniformly) distributed, but the news media says otherwise.
The Georgia department of Public Safety published records for the month of September.
These records indicated the point at which 30 accidents occurred, as follows:

88.3 40.7 36.3 27.3 36.3

91.7 67.3 7.0 45.2 23.3
98.8 90.1 17.2 23.7 97.4
32.4 87.8 69.8 62.6 99.7
20.3 73.1 21.6 6.0 45.3
76.6 73.2 27.3 87.6 87.2

Use the Kolmogorov – Smirnov test to determine whether the distribution is uniformly
distributed given D0.25, 30=0.24

4. (a) Explain time series input models with suitable examples.

(b) The following data were available for the past 10 years on demand and lead time.

Lead time: 6.5 4.3 6.9 6.9 6.9 6.9 5.8 7.3 4.5 6.3
Demand: 103 83 116 97 112 104 106 109 92 96

Estimate correlation and covariance.

5. (a) Explain in detail the 3-step approach of Naylor and finger in the validation process.
 138

(b) Mention some important points which you would consider in selecting simulation
software.

Write a program for single server queue using any one of the simulation language you
know.

6. (a) Explain point estimation and interval estimation of O (or ).

(b) Discuss the output analysis for terminating simulations and confidence interval
estimation for a fixed number of replications.

7. What are the objectives of simulation in a manufacturing system? Give the block
diagram and explain the sequence of operation in a manufacturing system. Suggest a
suitable simulation language for the same.


