Queuing Systems

Waiting for services is part of our daily life. We wait to eat in restaurants, we "queue up" at the
checkout counters in grocery stores, and we "line up" for services in ARMs. And the waiting phenomenon
is not an experience limited to human beings only: Jobs wait to be processed on a machine, planes circle
in a stack before given permission to land at an airport, and cars stop at traffic lights. Unfortunately, we
cannot eliminate waiting without incurring inordinate expenses. In fact, all we can hope to achieve is to
reduce its adverse impact to tolerable levels.

Why Study Queues?

The study of queues deals with quantifying the phenomenon of waiting in lines using
representative measures of performance, such as average queue length, average waiting time in queue,
and average facility utilization.

Waiting lines occur when there is a temporary imbalance between supply (capacity) and demand.
Waiting lines add to the cost of operation and they reflect negatively on customer service, so it is
important to balance the cost of having customers wait with the cost of providing service capacity.
Customer waiting lines occur when there is too little capacity to handle demand but having more capacity
than what is needed to handle demand means that there is idle (unproductive) capacity. From a
managerial perspective, the key is to determine the balance that will provide an adequate level of service
at a reasonable cost.

Waiting lines abound in all sorts of service systems. They are non-valued-added occurrences. In
lean systems, waiting is one of the eight wastes. For customers, having to wait for service can range from
being acceptable (usually short waits), to being annoying (longer waits), to being a matter of life and death
(e.g., in emergencies). For businesses, the costs of waiting come from lower productivity and competitive
disadvantage. For society, the costs are wasted resources (e.g., fuel consumption of cars stuck in traffic)
and reduced quality of life. Hence, it is important for system designers and managers of existing service
systems to fully appreciate the impact of waiting lines.

Elements of Queuing Model

The principal actors in a queuing situation are the customer and the server. Customers are
generated from a source. On arrival at a service facility, they can start service immediately or wait in a
queue if the facility is busy. When a facility completes a service, it automatically "pulls" a waiting customer,
if any, from the queue. If the queue is empty, the facility becomes idle until a new customer arrives.

From the standpoint of analyzing queues, the arrival of customers is represented by the inter-
arrival time between successive customers, and the service is described by the service time per customer.
Generally, the inter-arrival and service times can be probabilistic, as in the operation of a post office, or
deterministic, as in the arrival of applicants for job interviews

Queue size plays a role in the analysis of queues, and it may have a finite size, as in the buffer
area between two successive machines, or it may be infinite, as in mail order facilities.
 Queue discipline, which represents the order in which customers are selected from a queue, is
an important factor in the analysis of queuing models. The most common discipline is first come, first
served (FCFS). Other disciplines include last come, first served (LCFS) and service in random order (SIRO).
Customers may also be selected from the queue based on some order of priority. For example, rush jobs
at a shop are processed ahead of regular jobs.

The queuing behavior of customers plays a role in waiting-line analysis. "Human" customers may
jockey from one queue to another in the hope of reducing waiting time. They may also balk for joining a
queue altogether because of anticipated long delay, or they may renege from a queue because they have
been waiting too long.

The design of the service facility may include parallel servers (e.g., post office or bank operation).
The servers may also be arranged in series (e.g., jobs processed on successive machines), or they may be
networked (e.g, routers in a computer network).

The source from which customers are generated may be finite or infinite. A finite source limits
the customers arriving for service (e.g., machines requesting the service of a repairperson). An infinite
source is forever abundant (e.g., calls arriving at a telephone exchange).

Generalized Poisson Queuing Model

The development of the generalized model is based on the long-run or steady-state behavior of
the queuing situation, which is achieved after the system has been in operation for a sufficiently long time.
This type of analysis contrasts with the transient (or warm-up) behavior that prevails during the early
operation of the system. One reason for not discussing the transient behavior in this chapter is its
analytical complexity. Another reason is that the study of most queuing situations occurs under steady
state conditions.

The generalized model assumes that both the arrival and departure rates are state dependent,
meaning that they depend on the number of customers in the service facility. For example, at a highway
toll booth, attendants tend to speed up toll collection during rush hours. Another example occurs in a
shop with a given number of machines where the rate of breakdown decreases as the number of broken
machines increases (because only working machines are capable of generating new breakdowns).

Defined as

n = Number of customers in the system (in-queue plus in-service)

λn = Arrival rate given n customers in the system

μn = Departure rate given n customers in the system

pn = Steady-state probability of n customers in the system

 Figure 1 Poisson queues transition diagram

The generalized model derives p, as a function of λn and μn. These probabilities are then used to
determine the system's measures of performance, such as the average queue length, the average waiting
time, and the average utilization of the facility.

The probabilities pn are determined by using the transition-rate diagram in Figure 1. The queuing
system is in state n when the number of customers in the system is n. The probability of more than one
event occurring during a small interval h tends to zero as h → 0. This means that for n > 0, state n can
change only to two possible states: n - 1 when a departure occurs at the rate μn , and n + 1 when an arrival
occurs at the rate λn. State 0 can only change to state 1 when an arrival occurs at the rate λ0 . Notice that
μ0 is undefined because no departures can occur if the system is empty.

Specialized Poisson Queues

Figure 2 depicts the specialized Poisson queuing situation with c parallel servers. A waiting
customer is selected from the queue to start service with the first available server. The arrival rate at the
system is λ customers per unit time. All parallel servers are identical, meaning that the service rate for
any server is μ customers per unit time. The number of customers in the system is defined to include those
in service and those waiting in queue.

Figure 2 Schematic representation of a queuing system with c parallel servers

 A convenient notation for summarizing the characteristics of the queuing situation in Figure 2 is
given by the following format:

(a/b/c):(d/e/f)
where
a = Arrivals distribution
b = Departures (service time) distribution
c = Number of parallel servers (= 1,2, . ,00)
d = Queue discipline
e = Maximum number (finite or infinite) allowed in the system (in-queue plus in-service)
f = Size of the calling source (finite or infinite)

The standard notation for representing the arrivals and departures distributions (symbols a and
b) is

M = Markovian (or Poisson) arrivals or departures distribution (or equivalently exponential inter
arrival or service time distribution)

D = Constant (deterministic) time

E = Erlang or gamma distribution of time (or, equivalently, the sum of independent exponential
distributions)

GI = General (generic) distribution of inter- arrival time.

G = General (generic) distribution of service time

The queue discipline notation (symbol d) includes

FCFS = First come, first served

LCFS = Last come, first served
SIRO = Service in random order
GD = General discipline (i.e., any type of discipline)

To illustrate the use of the notation, the model (M/D/10):(GD/20/00) uses Poisson arrivals (or
exponential inter-arrival time), constant service time, and 10 parallel servers. The queue discipline is GD,
and there is a limit of 20 customers on the entire system. The size of the source from which customers
arrive is infinite.

As a historical note, the first three elements of the notation (a/b/c), were devised by D.G. Kendall
in 1953 and are known in the literature as the Kendall notation. In 1966, A.M. Lee added the symbols d
and e to the notation. This author added the last element, symbol f, in 1968. Before presenting the details
of the specialized Poisson queues, we show how the steady-state measures of performance of the
generalized queuing situation can be derived from the steady-state probabilities.
Steady-state Measures of Performance

The most commonly used measures of performance in a queuing situation are

Ls = Expected number of customers in system

Lq= Expected number of customers in queue

Ws =Expected waiting time in system

Wq = Expected waiting time in queue

Po = Probability that no customers are in the queuing system

Pn = Probability that n customers are in the queuing system

ρ = Probability that the server is busy (utilization factor)

*Recall that the system includes both the queue and the service facility

Single-server Model (M/M/1):(FCFS/∞/∞)

A single-server with a single-waiting line is the simplest form of queuing system. The following are
the assumptions of the M/M/1 model:

1. An infinite calling population

2. A first-come, first-served queue discipline

3. Poisson arrival rate

4. Exponential service times

These assumptions have been used to develop a model of a single-server queuing system.
However, the analytical derivation of even this simplest queuing model is relatively complex and lengthy.
Thus, we will refrain from deriving the model in detail and will consider only the resulting queuing
formulas.

Given that

λ = the arrival rate (average number of arrivals per time period)

μ = the service rate (average number served per time period)

and that λ < μ (customers are served at a faster rate than they arrive, we can state the following
formulas for operating characteristics of a single-server model.

1. The probability that no Customers are in the queuing system (either in the queue or being served)
is
 𝜆
𝑃𝑜 = 1 −
𝜇
2. The probability that n customers are in the queuing system is
𝑛 𝑛
𝜆 𝜆 𝜆
𝑃𝑛 = ( ) • 𝑃𝑜 = ( ) (1 − )
𝜇 𝜇 𝜇

3. The average number of customers in the queuing system is

𝜆
𝐿𝑠 =
𝜇−𝜆

4. The average number of customers in the waiting line is

𝜆2
𝐿𝑠 =
𝜇(𝜇 − 𝜆)

5. The average time a customer spends in the total queuing system is

1 𝐿𝑠
𝑊𝑠 = =
𝜆−𝜇 𝜆

6. The average time a customer spends waiting in the queue to be served is

𝜆
𝑊𝑞 =
𝜇(𝜇 − 𝜆)

7. The probability that the server is busy (i.e., the probability that a customer has to wait), known as
utilization factor, is
𝜆
ρ=
𝜇

Multiple-Server Model (M/M/c):(FCFS/∞/∞)

Slightly more complex than the single-server queuing system is the single waiting line being
serviced by more than one server (i.e., multiple servers). A multiple-channel waiting line consists of two
or more channels or service locations that are assumed to be identical in terms of service capability. In
the multiple-channel system, arriving units wait in a single line and then move to the first available channel
to be served. Examples of this type of waiting line include an airline ticket and check-in counter where
passengers line up in a single line, waiting for one several agents for service.

The following are the assumptions of the M/M/c model:

1. The waiting line has two or more channels,

2. The arrivals follow a Poisson probability distribution,
3. The service time for each channel follows exponential probability distribution,
4. The mean service rate is the same for each channel
5. The arrivals wait in a single waiting line and then move to the first open channel for service,
and
 6. The queue discipline is first-come, first-served (FCFS)

The parameters of the multiple-server model are as follows:

λ = the arrival rate (average number of arrivals per time period)

μ = the service rate (average number served per time period)

c = the number of servers

cμ = the mean effective service rate for the system, which must exceed the arrival rate

1. The probability that no customers are in the queuing system (either in the queue or being served) is
1
𝑃𝑜 = 𝑛
1 𝜆 1 𝜆 𝑐 𝑐𝜇
[∑𝑐−1 (
𝑛=0 𝑛! 𝜇) ] + 𝑐! (𝜇) (𝑐𝜇 − 𝜆 )
2. The probability that n customers are in the queuing system is
𝑛
1 𝜆
𝑃𝑛 = ( ) 𝑃𝑜, 𝑓𝑜𝑟 𝑛 > 𝑐
𝑐! 𝑐 𝑛−𝑐 𝜇
1 𝜆 𝑛
𝑃𝑛 = ( ) 𝑃𝑜, 𝑓𝑜𝑟 𝑛 ≤ 𝑐
𝑛 𝜇
3. The average number of customers in the queuing system is
𝜆 𝑛
𝜆𝜇 ( ) 𝜆
𝜇
𝐿𝑠 = 2
𝑃𝑜 +
(c − 1)! (c𝜇 − 𝜆) 𝜇

4. The average number of customers in the waiting line is

𝜆
𝐿𝑞 = 𝐿𝑠 −
𝜇

5. The average time a customer spends in the total queuing system is

𝐿𝑠
𝑊𝑠 =
𝜆
6. The average time a customer spends waiting in the queue to be served is
𝜆 𝐿𝑞
𝑊𝑞 = 𝑊𝑠 − =
𝜇 𝜆

7. The probability that a customer arriving in the system must wait for service (i.e., the probability that
all the servers are busy) is
1 𝜆 𝑐 𝑐𝜇
𝐿𝑠 = ( ) ( ) 𝑃𝑜
c! 𝜇 𝑐𝜇 − 𝜆