Introduction

Queueing is an aspect of life which we encounter regularly in our daily

lives. It happens when we checkout at the supermarket, when we log onto

the internet, when we are stopped at a traffic light, and even when we want

to ride the newest roller coaster. But what exactly is queueing theory? A

queue is an organized sequence of items such as data, messages, people,

cars, etc. A queue may also be referred to as a line. Queueing theory is the

study of these organized sequences of items from a mathematical

standpoint. More specifically differential calculus, integral calculus,

probability, differential equations, and some matrix manipulation techniques

are requisites to analyze systems through the use of queueing theory. Since

queueing is a consistent facet of our everyday lives, it seems logical to

further study the theory of queueing.

Why would we want to better understand queueing theory? Intuitively,

the better we understand a system, the greater the likelihood of improving

the system. But why improve the system? Improving or increasing efficiency

of a system will generally benefit all who is a part of that particular system.

For example, consider a congested runway-landing at an airport. This would

create several airplanes to be circling around the airport in a holding pattern

while they wait their turn to land, causing problems. The needless burning

of excess jet fuel could cost the airline a considerable amount of money

which translates into higher ticket prices. Also the excess fuel usage emits

1|Jaime
more pollutants than necessary into the atmosphere, which could possibly

create undesirable working or living conditions. Also consider a gridlocked

street where all cars on both sides of the street are at a standstill. One

obvious problem is sitting in your car for a copious amount of time which will

delay your destination arrival. Another possible dilemma could be that an

emergency vehicle will not be able to pass through the street, causing them

to take a detour delaying their arrival, which could drastically alter the state

of the emergency for which they were called. For a more superficial example,

consider a world without drive-through service at all fast-food restaurants.

Queueing theory allows the restaurant to develop the necessary protocols to

service their consumers without leaving their vehicle. Queueing theory also

allows the restaurant to service all consumers faster and more often,

benefitting both parties. The restaurant is able to make more money by

servicing more customers in a shorter amount of time and the customer has

an improved quality of dining by not having to vacate their vehicle or wait too

long to receive their food. There are many other reasons why the

optimization of an everyday system is essential.

Six Model Characteristics

How do we improve a system? We must first create a model by using

quantifiable characteristics. There are six basic characteristics of processes

that can be used to provide an adequate description of a queueing system

2|Jaime
(depending on the complexity of the system, other characteristics may be

added or removed).

1. Input

2. Service

3. Queue discipline

4. System capacity

5. Number of service channels

6. Service stages

Input

This pattern, also known as the input, is generally measured by the

average amount of arrivals to the queue per unit of time (average arrival

rate). It could also be measured by the average time between successive

arrivals (average interarrival time). Either one of these measures are

adequate in describing the system input because the quantities are visibly

related. If there is certainty in the input, then the arrival pattern can be

accurately determined by one of the two measures. If there is uncertainty in

the arrival pattern, then the two measures only provide a central tendency of

arrival. When this happens, a probability distribution corresponding to the

randomness of the arrival pattern will be required (Gross, p.2).

A major point of interest concerning the input process is the different

sizes of arrival into the system. If more than one person can enter into the

queue concurrently, then the input is said to occur in bulk or batches. Note

3|Jaime
that multiple people entering the queue concurrently does not mean they all

will receive service at the same time (Gross, p.3).

The customer’s reaction is vital knowledge to the system. Impatient

customers engage in three different behavior patterns. A person may either

stay in the queue or leave the queue upon arrival. The term balked is used to

describe a person that does not enter the queue upon arrival. A person is

said to have reneged if (s)he enters the queue, then leaves due to lack of

patience. A person jockeys for position if (s)he moves from one line to the

other if given two or more parallel waiting lines (Gross, p.3).

The last factor to be taken into consideration is the way in which the

pattern changes over time. A stationary arrival pattern is one which does not

change over time. A nonstationary arrival pattern is one that is not time-

independent (Gross, 3).

Service

A service pattern can be classified by the number of people served per

unit of time, or the time required to service the person. Even though the

service patterns are similar to arrival patterns with respect to rate, there is a

vital difference between the two. The service rate and service time are

conditioned on the fact that system is not empty. Therefore, there exists at

least one person in the queue in need of service. If the queue is empty, the

system is considered to be idle. Service, like the arrival, can be deterministic

4|Jaime
or probabilistic and require the corresponding probability distributions

(Gross, p.3).

Service, like arrival, can occur one at a time or in bulk. It is easy to

consider the case in which people are serviced one at a time. In the latter

case we should consider the service of customers simultaneously with a

single server. For example, when multiple internet users visit the same

website and make purchases concurrently. Another instance is when many

people board a train or a bus to travel (Gross, p.3).

If we are considering a service in which humans are servicing and

arriving, we must take into consideration the human factor. A server may

increase the rate of service as a result of the queue building up. Conversely,

the server may become disconcerted and therefore become inefficient. The

kind of service can be classified as state-dependent service because the rate

of serviced is determined by the amount of people in the queue (Gross, p.4).

Much like the arrivals, service can be stationary and nonstationary with

respect to time. Consider the case where there is a cashier being trained

while servicing a customer. Efficiency will eventually increase as the more

experience is gained. This is similar to state-dependent service, but

different. Note that any system can be state-dependent and nonstationary

simultaneously (Gross, p.4).

It is pretty much inevitable that customers will still have to wait in line

no matter how fast the rate of service is. Generally, customers arrive and

5|Jaime
depart at irregular intervals, which means that there will be no definitive

pattern. Therefore a probability distribution for queue lengths would be a

result of arrivals and services which are generally assumed to be mutually

independent (Gross, p.4).

Queue Discipline

This refers to the selection method of customers for service. There is

the first come first serve (FCFS) which is also known as first in first out (FIFO).

This is the most common selection method that we see in our everyday lives.

Another discipline is called last come, first served (LCFS). This is most

common for inventory systems where it is easier to reach the last item that

was stocked. Random selection for service (RSS) is the order independent of

time of arrival to the queue (Gross, p.4).

There are two kinds of priority disciplines: preemptive and

nonpreemtive. With the preemptive priority, the customer with the highest

priority moves to the front of the queue and receives service immediately and

the customer with the lowest priority is stopped mid-service. The

nonpreemptive is when the person with the highest priority moves to the

front of the queue but does not get serviced until the customer with the

lower priority is finished being served (Gross, p.5).

It is possible to have preemptive and nonpreemptive disciplines

concurrently. To do this we must assume two things. First, we need to use

the FCFS discipline. Second, we must list a set of priorities from least to

6|Jaime
greatest, say 1-4. Say that there are several customers with priority 3. They

will be served in the FCFS order nonpreemptively until they are halted by a

customer of priority 4. Then the priority 3 customer being serviced will be

preempted (Gross, p.5).

System Capacity

Some queueing systems have limited space to hold all customers

demanding service. This is generally referred to as finite queueing

situations. This essentially forces customers to balk, due to the limited

space of the system’s location (Gross, p.5).

Number of Service Channels

The numbers of service channels refer to the number of servers that

can simultaneously service more than one customer. This can happen one of

two ways. The first way is a single queue with multiple servers. The second

is a set of multiple queues with multiple servers. In these cases, it is

generally assumed that each parallel channel operates independently of one

another (Gross, p.6).

Service Stages

Most queueing systems that we are familiar with have a single stage

like the checkout counters of a grocery store. Others have multiple stages

such as an assembly line. After an item runs through several stages of

assembly it arrives at a quality inspection station in which it can be sent to

7|Jaime
the back of the queue for reassembly due to inferior construction (Gross,

p.6).

A Simple Queueing Model

Now given our six characteristics of a model, we can create a simple

model to better understand how they work. To construct this simple model

we must first make a lot of simple assumptions. Consider a queue with only

one queue, one server, and an infinite number of waiting positions.

 Let the average arrival rate be denoted by 𝜆.

 Let the time interval be denoted by Δ𝑡.

 Let the average service rate be denoted μ.

 Let 𝑝𝑁 𝑡 = 𝑃(𝑠𝑦𝑠𝑡𝑒𝑚 𝑖𝑛 𝑠𝑡𝑎𝑡𝑒 𝑁 𝑎𝑡 𝑡𝑖𝑚𝑒 𝑡)

Now as Δ𝑡→0, we have some probabilities to consider.

1. P(one arrival in time interval Δ𝑡)= 𝜆Δ𝑡

2. P(no arrivals in time interval Δ𝑡)= 1 − 𝜆Δ𝑡

3. P(one departure from system in time interval Δ𝑡)= 𝜇Δ𝑡

4. P(no departures from system in time interval Δ𝑡)= 1 − 𝜇Δ𝑡

The system state at any time instant is the number in the system at that

instant. Note that the number in the system includes those waiting, and

those being serviced. The probability of the system state at time 𝑡 + Δ𝑡 is

8|Jaime
 𝑝0 𝑡 + Δ𝑡 = 𝑝0 𝑡 1 − 𝜆Δ𝑡 + 𝑝1 𝑡 𝜇Δ𝑡 for N=0

𝑝𝑁 𝑡 + Δ𝑡 = 𝑝𝑁 𝑡 1 − 𝜆Δ𝑡 − 𝜇Δ𝑡 + 𝑝𝑁−1 𝑡 𝜆Δ𝑡 + 𝑝𝑁+1 𝑡 𝜇Δ𝑡 for N>0

and they are subject to the normalization condition, that is, the sum of all

system state probabilities with respect to time is equal to one.

𝑝𝑖 𝑡 = 1 ∀𝑡 ≥ 0
∀𝑖

Taking the limits as Δ𝑡→0 subject to the same normalization condition gives

𝑝0′ 𝑡 = −𝜆𝑝0 𝑡 + 𝜇𝑝1 𝑡 for N=0

𝑝𝑁′ 𝑡 = − 𝜆 + 𝜇 𝑝𝑁 𝑡 + 𝜆𝑝𝑁−1 𝑡 + 𝜇𝑝𝑁+1 (𝑡) for N>0.

These differential equations, subject to the normalization condition, yield the

transient and equilibrium solutions (Bose, p.3-4).

Little’s Forumla

As a byproduct of a very simple queue model is called Little’s Formula

9|Jaime
 𝐿 = 𝜆𝑊

where L is the long run average of the queue length, W is the waiting time

plus the service time, and λ is the arrival rate of the customers. Note that

these quantities are assumed to be finite (Prabhu, p.9).

Conclusion

This paper was simply a brief introduction into queueing theory. I have

discussed what a queue is and how we define queueing theory. I have also

discussed six characteristics that make up a model and the traits of each

characteristic. I presented a few differential equations, exploiting Little’s

Formula, to illustrate a simple way of creating a queue model based on a

rudimentary system. Since queueing theory has many applications, there are

many ways to model a system.

10 | J a i m e
 REFERENCES

Bose, K. Sanjay An Introduction to Queueing Systems Kluwer

Gross, Donald Fundamentals of Queueing Theory John Wiley &

Prabhu, N.U. Foundations of Queueing Theory Kluwer Academic

11 | J a i m e