The problem with

waiting time
Why the only way to real optimization of any process requires discrete event simulation

Bill Nordgren, MS CIM, FlexSim Software Products

ver the years there have been many Where does waiting time or queuing begin?

O management techniques introduced to

reduce and manage the waste in a pro-
cess. These include: just-in-time, Kanban,
Between each transformation step in any process is a
period in which the item being produced must wait.
business process re-engineering, lean manu- This time will be defined as the product waiting time.
facturing, and six-sigma. There is no doubt There are cases when waiting is the objective, such as
that these techniques and processes are use- the money supply of a bank: money held overnight or
ful when it comes to efficiency and decisive- for a period of days pays interest. But in most cases,
ness. However, the most important factor, waiting time and queuing is a recurring evil that can
profitability, which should be considered first, suck away the profits in a process. The fact is that
is often lost in the process because most engi- waiting time can comprise 90 percent of the total
neers are not directly responsible for company throughput time of a process. If this seems high, take
profits. a walk through your production facility and measure
These management techniques have one goal: to the time product sits around versus the time it is in
reduce the waiting time (waste) in a process. Waiting actual production or being worked on. You will see
time in a process can be grouped into three categories: very quickly the seriousness of the problem.
customer waiting time, product waiting time, and Waiting time occurs in every industry, including
machine waiting time. The problem with these wait- manufacturing, warehousing, supply chain, services,
ing times is that they result in a tug of war that pits and healthcare. When was the last time you were
customers, products, and machines against each other. annoyed by waiting in the line at the supermarket
The difficult question to answer is what kind of wait- or delays at the doctors office? Although difficult
ing time reduction saves the most money. To get an to quantify in terms of money, these waiting times
understanding of the difficulty to solve the question create irritation and inconvenience for customers that
at hand you have to understand the dynamics of the can be qualitatively high.
process itself and how the three types of waiting time In many cases the translation from the physical
impact the total cost of the process. world into a queuing model is made easy by analyzing
the constraints (processes with high utilization) of a

Page 1 of 8
system. In other cases the translation is not so evident. possible utilization. Because the financial side of a
For example it may be easy to calculate the cost of business does not understand the effects high resource
floor space used to queue parts in production, but it utilization (95 percent load factor) has on the system
is another matter to calculate the cost of customers as a whole, process efficiency may be doomed even
who have to wait for service that might never return before the first product rolls off the line.
because of the frustration of waiting.

Let the tug of war begin

The cost of waiting time
Every process or logistics problem can be viewed as
Every process system contains the three types of wait- a distribution problem of the cost of waiting times.
ing time. If you understand and are able to weight Factors that impact these waiting times include but
the cost of each type of waiting time, it is possible to are not limited to variability, quality, equipment relia-
determine the optimum cost of the total waiting time. bility, staff reliability, and system load factors. When
Lets take a look at each type and weight the cost of each type of waiting time increases or decreases it
each. has an impact on the other; this creates a tug of
war between the costs of waiting. For example, if
Customer waiting time you wanted to satisfy the demand of every customer
immediately you would have to maintain a finished
The customer is king. Without customers you have no goods supply of every product you make. The cost of
business, and in every business the goal is to service product waiting would sky rocket. On the other hand,
the customer as quickly as possible. If a customer has if you wanted to reduce your work in process you run
to wait for products or services they will always look the risk of not having product at the right place at
for another source that can service them in a timely the right time; your machines may be idle waiting
manner. As a result, the cost of customer waiting for product, thus increasing the cost of waiting for
time is the most expensive. If you have no customers machines. This tug of war is not just between two
the amount of waiting time for products or machines sidesit is a 3-way war. Each can affect the other
is moot. two.
For example, a machine that is loaded too high will
Product waiting time cause a large queue to build in front of the machine.
It may also starve those processes downstream. Thus,
Product waiting time includes all the waiting time a the product waiting cost increases and deliveries to
product (raw material, work-in-process, and finished the customer may fall behind, increasing the cost of
goods) spends in all its different states. This waiting customer waiting time. This complex dance of wait-
time will be weighted as the second most expensive ing times can be difficult to understand and control
waiting time of the three. When you consider the unless you know how to analyze and experiment with
cost of raw material storage and finished good storage the correct variables. Before we look at the tools
(warehouses) you quickly see the cost involved. This used for such analysis and experimentation, we need
cost or waste as seen by the lean engineer is the main to understand the dynamic behavior of what causes
target for kaizen events. waiting time.

Machine waiting time

How does a queue start?
Machine waiting time is the time a resource (machine
or human) is idle waiting to work on a product. This Queues start the moment the supply of products or
cost is ranked number 3 because it relates only to customers is greater than the processing capacity. As
the cost of an idle resource. It can be costly if the a result, an unstable situation develops and, theoreti-
resource is very expensive to purchase. The ROI of cally, the queue will increase indefinitely as the utiliza-
equipment is always scrutinized by the financial side tion of the system approaches 100 percent. Queues
of the business to insure equipment has the highest build when supply exceeds capacity and decrease when

Page 2 of 8
supply is less than capacity. Another factor that can of variability in arrivals or process time. By reduc-
influence the queue is variation in the processing time.ing both factors, the waiting time of the product or
We all know what it is like to get behind the person in customer can be reduced. However, when you only
the checkout line who has all kinds of special requests reduce machine utilization, machine waiting time can
that cause you to wait endlessly. When viewed over become a factor. In a system where you have a large
a long period of time the average utilization of a ma- degree of variability, increasing capacity is the only
chine or resource may be less than 100 percent, but way to reduce the utilization of a machine and the
variation can cause queues to build temporarily or number in the queue.
you could have a situation where machines are starved Many processes show a lot more variation than you
for a period of time. might think. Breakdowns, pauses, re-sets, waste, oper-
The equation for the calculation of average waiting ator error, and restocking make a seemingly constant
time (Wt ) is shown in 1; the relationship between process extremely variable. Over time the process
utilization () and variability is expressed in terms might seem constant, but on a weekly, daily, or hourly
of the variance coefficient (cv ) and waiting time (Wt ).
basis production has a very stochastic nature. With
Average processing time is shown as (Pt ). the numerous uncertainties, one can assume your sys-
tem is a Poisson system and will have a variance
Wt = 0.5(1 + cv )( )(Pt ) (1) coefficient of one. Figure 2 shows the relationship of
1
utilization and waiting time of products and machines.
When the utilization of a machine increases to more
than 70 percent and there is a high degree of variation
in arrivals, a situation can develop where a queue can Figure 2: The relationship among utilization and waiting
build. The queue can only be eliminated if arrivals are times with a Poisson process.
limited or there is added capacity. The relationship
between utilization, variability, and average waiting
of the product or customer is shown in Figure 1.

Figure 1: The relationship between utilization and

variability.

Note that cv = Average .

By adding the waiting time of the products and

machine you can find the optimum utilization for the
machine. The minimum cost is derived by calculating
the weighted cost factors related to each waiting time.
The exponential form of the waiting time graph often
confines the optimum utilization of a machine between
A constant arrival pattern has a variance coefficient 70 and 90 percent.
of zero. An independent arrival pattern (Poisson dis-
tribution) has a variance coefficient of one. Figure Waiting Time Law: The higher the
1 shows that there is only a small area where long utilization of a machine or resource,
waiting times occur. Waiting time is only a prob- waiting time increases exponentially.
lem when there is high utilization and a large degree

Page 3 of 8
 0.9
Figure 3: Wt = 0.5(1 + 1)( 10.9 )(5) = 45 minutes.
When you consider the dynamics of utilization,
waiting time, and the cost of waiting, you can see the
Process Time
inherent problems designed into systems where equip-
12%
ment is purchased with an ROI basis of 90 percent (or
88%
higher) load factor. The process of buying equipment
already has built in bottlenecks and waste that cost Waiting Time
real money and time on the production floor.

duce the time by 10 percent (30 seconds), would the

Minimizing queues average waiting time be reduced significantly? At first
look, it doesnt appear that the waiting time would be
Queuing problems develop from a combination of high
reduced by much. However, as a result of the decrease
utilization and variability in the arrival pattern and
in processing time, the utilization would drop to 81
the process time of a machine. Another factor that
percent. Since waiting time increases exponentially
can cause queuing is batch processing, which increases
with utilization, the waiting time could be reduced
the process time because the batch cannot move un-
significantly (Figure 4). With this value, the waiting
til the last piece in the batch is completed. Batch
time is 19 minutes. The total throughput decreases
processing is used to minimize the movement steps be-
from 50 minutes to 23.5 minutes, a reduction of ap-
tween processes and setup times at the process. Using
proximately 50 percent. This seemingly small change
techniques to minimize the setup time when chang-
in processing time results in a significant reduction of
ing part types has reduced the variation in batching
waiting time.
processes. Optimization of scheduling batches that
reduce setup based on similar part makeup is vital to
prevent unnecessary delays in processing. 0.81
Figure 4: Wt = 0.5(1 + 1)( 10.81 )(4.5) = 19 minutes.
There are three ways to add capacity to a process Process Time
that will help reduce the utilization of the process
and the product waiting time. The first is to reduce
the stochastic nature of the arrival rate of products, 48%
the second would be to add a machine, and the third
is to reduce the process time on the machine. It 52%
is not easy to fix the stochastic nature of arrivals;
adding expensive machines or resources may not be Waiting Time
cost effective. There is a persistent notion that a minor
reduction in process time has no significant influence
Using math to calculate waiting time is easy for sim-
on waiting time and throughput. The reasoning is
ple M/G/1 systems. What about more complex sys-
that process time only accounts for 5 to 10 percent of
tems that are commonplace in todays manufacturing
the total throughput time. However, this reasoning
and logistic environment? Mathematical analysis and
ignores the relationship between the utilization of a
spreadsheet calculations fall far short of measuring
machine and the wait time in the queue associated
and understanding the behavior of complex stochastic
with the machine.
processing and logistic systems. Understanding and
Suppose the processing time of a production step is
managing the competing costs of waiting is critical to
five minutes and the utilization of the machine is 90
the profits of every company, yet very few understand
percent. Using the Pollaczek-Kyntchin equation for
the relationship and effect that waiting has on costs
M/G/1 queuing systems, one can calculate that the
and customer satisfaction, let alone have any idea
average waiting time is 45 minutes (Figure 3). The
how to measure the effects so they can manage the
total throughput time is 50 minutes, resulting in 90
process.
percent waiting time and 10 percent processing time.
If you could optimize the working process and re-

Page 4 of 8
Discrete Event Simulation The experiment

The only tool ever developed to analyze the stochastic To run the experiment we will look at two scenarios
nature and tug of war impact, relationship, and cost (see Figure 6).
of customer, machine, and product waiting time is
discrete event simulation. With the use of simulation
you dont have to guess how your system is going to
behave before or after change. Simulation allows you
to scientifically experiment with your system before (a) Performance measures
you make real changes so you know with certainty
the changes you make will save real money. How
would you like to know with 95 percent confidence
that you will save 22 percent of the cost of work-in-
process storage costs? Or, that you have 95 percent
(b) Scenario configuarion
confidence that you can reduce the delivery time of
a product to your customer by 65 percent? You can
with the use of simulation.

Proof waiting time exponentially increases

with machine utilization
Simulation can be used to prove waiting time expo- (c) Experiment configuration
nentially increases as machine utilization increases.
We will experiment with an M/G/3. A source will Figure 6: Simulation experiment configuration for two
generate the arrival of a box every 10 seconds ex- scenarios.
ponentially distributed (to provide some stochastic
behavior to the model). Boxes then move to a queue
before moving to one of 3 machines for processing.
Scenario 1: Baseline
The processing time at each machine will be set to 30
seconds to balance the system. Boxes are sent to a The baseline will establish the utilization percentage
sink after processing and leave the model. A picture of the machine and the average wait time of boxes
of the model is in Figure 5. in the queue with an arrival rate of 1 box every 10
seconds exponentially distributed (Poisson process,
Figure 5: M/G/3 queuing system. cv = 1) and a cycletime of 30 seconds at each machine.
We will run the model for 8 hours of production
and we will run the model 20 times and average the
results. Results will be based on a 95% confidence
interval. Performance measures will be the average
utilization for all machines, the average wait time of
boxes in the queue, and the maximum content in the
queue. Results of the baseline experiment are shown
in Figure 7.
Results show an average of 98.72% utilization at
each machine, an average wait time of 228.75 seconds
for each box in the queue, and the average maximum
content of the queue as 59.65. All averages are based
on a 95% percent confidence interval (the min and
max range and standard deviation is shown for each
average).

Page 5 of 8
 (a) State Pie average replication plot (b) State Pie average data summary

(c) Average wait time in queue data summary (d) Average wait time in queue replication plot

(e) Maximum content of queue replication plot (f) Maximum content of queue data summary

Figure 7: Results of the baseline run.

Scenario 2: Proposed change rises exponentially as machine utilization increases.

The baseline model shows a 90% waiting time versus
In Scenario 2, the utilization percent of the machines a 10% processing time. The proposed system reduced
will be lowered by a 10% reduction in cycletime. This the waiting time percent to about a 50-50 percentage.
will result in a 27 second process time at each machine.
The real story is to look at the average waiting
The model will be run again with the same settings as
time for the boxes in the queue for each scenario.
the baseline with the only change being the cycletime.
For the baseline model, where machine utilization
The results are displayed and compared to the baseline
was approaching 99%, the average waiting time was
run in Figure 8.
almost ten times the processing time for each box.
Results show a 10% reduction of utilization of each The proposed model, with the cycletime reduced by
machine as a result of the 10% reduction in cycle- 10%, reduced the machine utilization by 10% and
time. Baseline model showed a utilization percent also reduced the average waiting time for each box to
of 98.72% and the proposed model shows an 89.8% only 32.05 seconds. In Figure 9 you can see that the
average machine utilization. The average wait time relationship between machine utilization and waiting
of each box fell sharply from 228.75 in the baseline time in the queue is exponential.
model to 32.05 in the proposed model. The maximum
content of the queue also fell sharply from 59.65 in
the baseline model to 19.4 in the proposed model.
This data confirms the hypothesis that waiting time

Page 6 of 8
 (a) State Pie average replication plot (b) State Pie average data summary

(c) Average wait time in queue data summary (d) Average wait time in queue replication plot

(e) Maximum content of queue replication plot (f) Maximum content of queue data summary

Figure 8: Results of the proposed run compared to the baseline run.

Conclusion them at the machines? Add more uncertainty and

the waiting time will increase. Simulation is the only
tool that allows you to build a model, experiment,
As you can see, the use of simulation made it easy and make informed decisions before you implement.
to understand the relationship of machine utilization How do you know your lean initiatives will work or
and product waiting. Simulation provides an array save as much money as presented? Build a model of
of data and charting options to easily identify and the current and proposed system and find out.
experiment with bottlenecks (areas of high utiliza-
tion) to understand the best way to solve problems. Most organizations do not understand or measure
What we have looked at so far is very simplistic but their process to even know that they have a problem;
would have taken hours to calculate by hand or using nothing is done to improve the process other than
spreadsheets. The real benefit of simulation is the the most obvious. Sitting by and doing nothing may
ability to build models of real life complexities and cost millions each year. Its possible that a fourth
relationships that are impossible to analyze any other type of waiting time that costs your company money
way. For example, how would the system change if is waiting to use simulation. No matter what your
people were introduced to move the boxes and process process, simulation is the only tool that can help you

Page 7 of 8
 classes in the use of simulation software. He
Figure 9: The exponential relationship between machine
received a Bachelor of Science in Manufactur-
utilization and waiting time.
ing Engineering Technology, and a Master of
Science in CIM (Computer Integrated Manu-
facturing) from Brigham Young University.

make the right decision every time. The tug of war

created by the cost of waiting customers, products,
and machines is complex and sometimes very hard to
understand. Methodologies, techniques, and heuris-
tics all play a role in process and system optimization.
Simulation is the only way to know if those method-
ologies, techniques, and heuristics are of real value to
the bottom line of the organization. Simulation helps
you understand the behavior of current systems and
processes and predict with confidence what will hap-
pen when proposed changes are implemented, without
the risks of actual physical change. Said another way,
speculation dies when simulation is employed.

About the Author

ill Nordgren is the President and CEO

B of FlexSim Software Products, Inc. In

1998, he founded ProModel Corporation
and was Vice President until he left in 1992.
In 1993, Bill founded F&H Simulation, Inc.
(now FlexSim Software Products, Inc.) and
introduced Taylor II, Taylor ED, and FlexSim
into the market.
Bill has authored several papers dealing
with simulation project management and
queuing theory, co-wrote the textbook Ap-
plied Simulation: Modeling and Analysis us-
ing FlexSim, and has taught hundreds of

This document Copyright

c 2014 FlexSim Simulation Software, Inc. All rights reserved.

Page 8 of 8