QUEUE MANAGEMENT 01/03

What is a queuing system

Modelling through queuing theory supports companies in order to take better

operations decisions (number of resources, bottlenecks….).
We can try to evaluate what are the number of resources that are needed to manage
any kind of service.
We can model our system and try to optimize our system in order to address a certain
demand based on the resources that we have.
Companies can understand which are the criticalities and where they are concentrated
in order to provide a better service.
We can better analyse the waiting times of our customers or our products.

A queuing system is any system that is composed by one or more customers (of any
type, they can be products, machine or people), that are in a place that could be a
waiting room, a production department, waiting to be served by one or more
“servers”.
A queuing system is composed by one or more customers (at least one) waiting to be
served by one or more servers.
Some examples:
Customers’ examples:
o People waiting at the cashier’s desk for a bank operation
o Pieces waiting to be processed by a lathe in a job-shop
o Container waiting to be loaded on chassis
Corresponding servers:
o Cashiers
o Job-shop’s lathest
o Overhead travelling cranes’ array for the loading
So, we talk about both production companies and service companies.
Examples: vaccine centres, ER (pronto soccorso), airport, highway (casello stradale),
car assembly line.
The characteristic of different queues can be different in terms of configuration. For
instance, the vaccine queue and the assembly line are obviously completely different
in terms of configuration.

Scheme of a queue:

The system is composed by the queue and the

server.

Whenever we deal with a system, always we

generate a queue.
Example:
Bank historical data show that on Monday there
are on average 12 customers/hour at the only
counter. The counter operator is able to serve 10
customers/hour on average.
Is a queue going to form?
Yes, the queue is going to form.
Another example:
Bank historical data show that on Monday between 10 and 11 am there are on
average 10 customers/hour (1 customer every 6 minutes on average) at the only
counter The counter operator is able to serve 12 customers/hour on average (1
customer every 5 minutes on average).
Is a queue going to form?
Also in this case, the queue will form. We cannot be sure that we are going to have
exactly this numbers of people, because we are talking about an average number.
Also, we cannot for sure say that we are not going to have any queue, because we
can have some moments in which for instance the number of customers is higher, or
the operator is unavailable.

Whenever you have a service rate that is lower than the arrival rate (first example),
you are going to have for sure a structural imbalance: the server can’t satisfy the
demand.
Even though you try to solve the structural imbalance, so set the service rate higher
than the arrival rate, you may face incidental imbalances, that depends on the
variability (difference between the average value and the actual value) and the
uncertainty (difference between the expected value and the actual value).

 Structural imbalance:
If the saturation level is too high
(arrival rate higher than service rate)
you will have issues in treat the
demand.
x-axis: saturation level of the system.
y-axis: units waiting to be processed.
In D the saturation is reached, so
every customer that will arrive have
for sure to wait. D represent the
queue, because in D we reached the
saturation.

 Incidental imbalance:

We work usually based on average values, that support us to take the decisions. But
there is always a variability that need to be considered and managed.
The structural imbalance of a queueing system is due to the fact that the average
pace of customers’ arrival into the system may be greater than the average of the
server’s serving rate. Queues are a result, a symptom.

Whenever we look at a system, we have different types of characteristics that we can

look at, both for service and production companies.
- Cost: in case of production companies, we talk about capital opportunity cost,
in case of service companies, we talk about time opportunity cost (people that
wait to receive the service).
- Space: in the production systems, if we are not able to manage our system
properly, we need bigger warehouses. The same for service companies, where
we need bigger waiting areas.
- Decouplement: regarding the stocks, you can decouple the production system,
to try to give the same pace to the different machines; regarding the service
companies, you can create job division and specialization.
- Utilization level: if we have bigger WIPs, creating a sort of aligned pace of the
machines, you keep them always saturated; the same for the service.
- Coordination: if there are WIPs it is less needed scheduling, non ho capito un
cazzo di questa parte.
Queueing can support the activities of a company.

Different elements of a queuing system

1. Calling population: is composed by the customers/WIPs entering our system

that need to be managed. It’s the input of the system.
We can have two types of population:
- Finite: in this case, the potential number of new customers for the system is
significantly affected by the number of customers already present in the system.
In other words, this happens when you have already a lot of people in the
system and this actually affect the way in which the service will be delivered
(for instance arriving to a copier shared between three secretaries).
- Infinite: in this case, the number of clients in the system does not affect the
demand pace for the service made by new customers (for instance people arriving
at the ER).

2. Arrival process: the customer arrive, with a certain pace. We will have two
activities:
- Rejecting: for several reasons (for instance there are no conformity with the
requirements), the person that arrive in the system is rejected by the system, and
he needs to go to another queue or another system.
- Balking: when there is a long queue, the customer may decides to go to another
system because of the high waiting time.

The arrival process can be described by a statistical distribution (negative

exponential distribution), which describes the time interval accruing between two
consecutive arrivals.

Example:
The average inter-arrival time between 2 patients to the doctor is 2.4 minutes.
Considering that a patient is just arrived, what is the probability that another
will arrive in the next 5 minutes?

So, knowing the arrival rate, we can compute the probability of having another
customer in a certain time period.

The Poisson distribution describe the number of events that can occur in a given
time t. This distribution gives the probability that n arrivals will arrive in a certain
moment during a certain interval. This probability is equal to:

Example:
At a telephone exchange phone, calls arrive at a pace of 12 calls per hour.
Knowing that the number of calls can be shaped with a Poisson distribution,
answer the following questions.
1) What is the probability that 10 calls will arrive in the next hour?
 2) What is the probability that there won’t be any call in the next 5 minutes?

3) What is the probability to have more than 2 calls in the next 10 minutes?

-> = 1-0.135-0.27-0.27

3. Queue configuration: if the customer decides to stay in the queue, maybe he

expected to wait 20 minutes, but is already 30 minutes it is waiting for the
service, he decides to go away (RENEGING).

4. Queue discipline: the queues are configurate, the people decide to stay in one
queue rather than another within the system. But then they see that another
queue is faster than the one that they have chosen and so decide to change
queue (JOCKEING).

5. Service process: the service is provided to the customer.

The service process describes how each server delivers the service, in particular
it defines its duration. It’s defined in terms of service times distributions.

We can use also for the service rate the negative exponential distribution.
Exercise:
Pierre Cahon, barman of one of the most popular discotheques of St. Catherine
Street, Montreal, serves cocktails at a pace of 1 every 50 seconds. Recently,
during an evening when the club was particularly full, every 55 seconds
someone was at the bar asking for a drink.
How to shape this system?

Relevant parameters:

With ‘q’ we are looking at the part of the system that is waiting in line, in the
queue.
With ‘s’ we are looking at the entire system, so people that are waiting in line
plus those who are made the server. So Ws, for instance, is the waiting time in
the queue + waiting time for the delivery of the service.

System utilization rate = saturation level = p (rò)

s = number of servers
(da non ricordare, date all’esame)
p (rò) = saturation level
P(n>K) = probability that there are more than n customers in the system

[formula valid for systems with 1 server (?)]

Example:

s=1

Nota: FCFS = first customer first served.

Example:

With the computation of the saturation level, we can understand how many
servers are needed at least to balance the arrival rate with the service rate. To
avoid a structural imbalance, we need at least 9 servers.

We want to give a certain service level to the customer: max waiting time in
queue 15 minutes.
So we compute the Wq, to understand how many servers we need to rech
Wq<15 minutes (at least 9 servers).
Wq= waiting time in the queue = saturation level / service rate – arrival rate
(da foglio formule)

With 12 servers we can reach a waiting time in queue lower than 15 minutes.
In our new system we will have 12 counters.
 OSS:

So, every time, you need to balance the cost of having an additional server (or
resource), with the benefits.

IMPORTANT: in every exercise, we must align the unit of time!!

6. Final step: according to what happened in the service process, the customer
may decide to repeat again the process (so return to calling population) or to not
use again the service in future because actually he was not satisfied for instance.

Sum up of the possible events in a system:

- Rejecting: when a customer is rejected by the system because she doesn’t respect
acceptance requirements
- Balking: when a customer decides to not enter a queue because it’s already too
long
- Reneging: when a customer already in queue gives up the service and goes away
without being served
- Jockeying: when a customer tries to trick the rules to get advantage, e.g. shifting
from one queue to another to shorten the waiting
The achievable goals when sizing a system

“Service quality” principles:

 Average waiting time of the customer

 Maximum waiting time

Cost principles:
 Minimization of the sum of the wait and service cost (trade off)

The choices of how many servers we will have

in the system should be taken also in base of
the variability present in the system.
If we have a very low variability, in most of
the cases the queue length won’t be so high.
If the variability is very high, we will have an
increase in the queue length.

When we want to keep the length of the

queue low, we can act in two ways.
- We can reduce the utilization level, increasing the number of resources.
- We can act on the variability reduction: we try to control the demand
(introducing a booking system to know when the people will arrive).
Queue configuration

We have different types of configurations of a queue.

 Singe queue: even though we have more than one
server, people are going to create one single queue. Is
based on the FIFO approach. This is used many cases
when the queue is very long and you can better manage
the line; they divide the overall demand arriving into the
system between all the servers.
Pros:
o Assure a FCFS type of service
o Avoid concern that another queue could be faster (you avoid the joking)
o Minimize the average waiting time
o Reneging actions are less frequent
Cons:
o It could “scare the customer”, so someone could not enter the system.

 Multiple queue: there are different lines and people

choose between the different queues. It can be seen as
a replication of a single system.
Pros:
o Allow to diversify the service (for instance, in a
certain queue you can serve pregnant woman, in
another queue you can serve people with less than 10 pieces)
o Allow to diversify the work
o Customer can chose a server of preference
o Balking actions are less frequent (because of quite small queue)
Cons:
o With the same number of serves, the average waiting time is greater

 Take a number queue: customers take the ticket,

don’t follow a specific line; FIFO approach.
Pros:
o Assure a FCFS type of service
o Avoid anxiety related to that another queue
could be faster
o Minimize the average waiting time
o Customers have the possibility to relax and dedicate themselves to other
thing during waiting time
Cons:
o An absent-minded customer could risk to loose its turn;
o It could “scare the customer” (if he/she doesn’t have anything else to do)

Examples:
Elements of choice between the types of queues:
High service customization degree  the best choice is a multiple queue
High service times variability  the best choice is a single queue

QUEUE CAPACITY:
- Limited queue
The number of customers in line plus the ones that are being served is limited.
Customers who arrive, after the capacity is overfilled, are rejected.
- Unlimited queue
The number of customers in line plus the ones who are being served is potentially
unlimited.

AT THE END OF THE DAY, summarizing:

Queue is formed by customers waiting to be served.

Queue discipline

Queue discipline (or “ranking rule”) is the rule or set of rules related to the order in
which customers are served, it’s often strictly related to the queue configuration.
Type of queue discipline:
 Static: the pertaining order among the customers in queue doesn’t change in time
and/or at the changing of the system conditions (ex. FCFS, SPT,…).
 Dynamic: the pertaining order between the customers can change in time and/or
at the changing of the system conditions. I change the way in which I serve my
customer, because of some emergency for instance.
What are the levers that I can use to manage queues and keep stable my system?
1. Levers and countermeasures offer side.
2. Levers and countermeasures demand side.
3. Levers and countermeasures for soft aspects/related to the psychology of
waiting.

 Offer side (service)

- Configuration changes
- Add/remove/move resources
- Decrease the service time, thanks to the support of technology or with training
of the operators
- Increase resources flexibility
- Decrease service times variability
 Demand side (arrivals)
- Decrease uncertainty level, introducing booking system or improving forecasts
- Decrease variance, with incentives for non rush hour moments or bookings
 Psychological lever
- We have to consider that the satisfaction = perception – expectation.
Perception is more important than reality.
- It’ s hard to play catch-up ball. First impression is the most important.
Measure the customer satisfaction is important to understand where the critical
points of the system are.

Psychological factors that may be considered:

- The unoccupied time seems longer than occupied time  distract and entertain
with small related or unrelated activities.
- Small show at Disneyland, mirrors near elevators, recorded music, magazines
at the doctor Pre-process wait seems longer than in-process wait  taking care
of the customer as soon as possible to make him/her “in-process”.
- Anxiety makes the wait to look longer  communicate and update frequently
the waiting time to the customer.
- An unfair wait seems longer than a fair one.
- The more the service is of value or the more the customer senses its utility, the
more he/she is willing to wait.
- Waiting alone seems longer than waiting in a group.
- A customer exposed to an exaggerated wait will be a more difficult one to serve
or an ex-customer.
Queue systems analysis tools 15/03

When we do the analysis of the queue systems, we might have three types of
approaches:
 Deterministic analysis
Pros: simple and intuitive to apply
Cons: doesn’t take transitory into account (the scenario is static); doesn’t take
arrivals and service times variability into account; queues are seen as an on-off
aspect (they are present or they are not).
 Queueing theory (approach that we’ll consider)
Pros: takes into account inter-arrivals and service times variability; allows to
compute a set of significant variables (Wq, Lq, Pn, etc…).
Cons: doesn’t take into account transitory (also in this case the scenario is static);
mathematic analysis can become very complex or even intractable in case of
complex waiting lines; non Poissonian arrivals or gaussian times (other times of
distribution are not taken into account); doesn’t consider customer complex
behaviours (balking, reneging, jockeying).
 Simulation
Pros: takes transitory into account; is very flexible.
Cons: time consuming; skills needed to design, production and analysis.

Queue theory assumptions:

- Stationary process: this assumption could be very restrictive: usually the arrival
process isn’t a static process, it changes during the day. For example, in a bank,
there are very different values of arrival process in a whole day  a possible
solution is to make more analysis considering different time slot (for instance you
can focus your attention of daily slots).
- Peculiar customers behaviours are not expected (no balking, reneging, jockeying).
- True only for certain inter-arrival and service times distributions.

Codification of the systems

We’ll use the Kendall’s codification to identify the type of system:

3 examples of codification of a system:

A queue system is formed by 1 queue and one or more servers to offer the service.
Therefore, hypothesizing that arrivals and service rates are deployed as negative
exponentials:

In the first case:

- M: negative distribution of customers’ arrivals
- M: negative distribution of service times
- 4: number of servers per queue
Some typical models of system:

The queuing theory is necessary for companies in order take better business decisions
in terms of resources.
Companies can better manage their resources and understand which is the need of
further resources in order to provide a service.
It studies the waiting times of customers/products.

EXERCISE 1.1
(M/M/c)

In the firsts morning hours customers arrive at the post office where there's a single
take a number queue, at an average pace of 54 customers/hour (Poisson) while each
server can manage to complete a service in an average time of 4 minutes (negative
exponential)
a) If there are 6 servers at the counter: what is the average number of customers
in the system, the average waiting time in line and in the system?
( Ls, Wq, Ws)
b) Determine the smallest number of counters that need to be opened to keep the
average time in the system lower than 10 minutes ( Ws<10 min)
c) If an employee costs 30€/h and customer's waiting time stands for a cost of
20€/h for customer, what would be the optimal number of servers according to
purely economic considerations?
Use the formulas in webeep pdf!
Solution:

Lq is obtained using the table:

(Lq obtained using the table):

Lq < 5.4; saturation = 3.6  c >= 5

We should minimize the sum of the cost related to the service provided and the
service delivered (every customer has an opportunity cost). We try to balance the
overall cost; everything is correlated to the number of servers that we put in the
system.
EXERCISE 1.2
(N systems M/M/1)

There are 3 systems M/M/1 (from the text).

Solution:
Ws customer type 1 = Ws photcopier + Ws cash counter
Ws customers type 2 = Ws fax + Ws cash counter
Ws tot = 2/3 * Ws1 + 1/3 * Ws2 = 10 minutes
 on average, the customer will take 10 minutes in the system.
Final considerations:
 A complex system is the composition of more elemental sub-systems that
interact among themselves (mapping the system and identifying every step of
the process).
 Very often in a complex system transit different type of customers (knowing in
advance every types of customers of one own system).
 The system expected throughput time is a representation of an "average"
customer and not of the specific one!
 The throughput time varies according to the variation of the work load in the
systems and at the variation of the input

EXERCISE 1.3
We need to use the node balancing rule:
Everything that enters in the node is equal to
everything that comes out from it. Nothing is created
or destroyed!!

Everything that enters in the system is equal to

everything that comes out. In the system there isn't
any flow loss. (The service process works on
customers. without creating nor destroying any!)

Solution:
We need to
- analyse the flows of the two types of products separately.
- analyse the flows in each of the nodes.
 16/03

“RINASIMENTO CLINIC” EXERCISE 2.1

Question 1:

Inactivity time: complementary value of the saturation: 1 – rò

3.Average lead time:

We need to consider the lead time of each type of customer (every type of customer
will wait a different amount of time).

-a che serve calcolare questi Wq se dopo calcoliamo Ws???

Occurrences computation (percentage of each type of customer on the overall number
of customers):

-perché si divide per 30???

% of customers type 1: customers that visit only Dr Carter, that don’t visit Dr
Romano.
% of customers type 2: customer that visit Dr Carter and Dr Romano.
% of customers type 3: customers that visit only Dr Carter.
The average lead time (or expected throughput time) will be the weighted sum of
the average lead time for each type of customer.

Notes:
Question 2:

We need to recompute the average Ws.

When we have MM1 system and we insert priority rule, we have two options of
priority:
 Pre-emptive priority:
If a customer of class 1 enters the system while a customer of class 2 is being
served, the service to the customer of class 2 is interrupted in order to give
immediate precedence to the customer of class 1.
The priority will be given immediately to customer of class 1, he have the
absolute priority.
Computation of Ws with pre-emptive priority:

 Non pre-emptive priority

If a customer of class 1 enters the system while a customer of class 2 is being
served, the service to the customer of class 2 is terminated and then the
customer of class 1 is taken care of.
Computation of Ws (system throughput time) with non pre-emptive priority:


Note: the alteration of the priority logic doesn't never change the system expected
throughput time (Ws); however, in some cases it's essential to increase the
customer's satisfaction degree and to improve customer service.

EXERCISE 2.2
Frame to solve the problem:
To compute the lambda we have to evaluate whether we have different type of
products in each system (in ophthalmology for instance we have special and normal
kits).

Now, since we have to compute the throughput time, we need to analyse the types of
customers. In this case we are going to meet 12 different typologies of customers:

We need to map our system, to observe in which stages every product will go and
then we can compute the frequency of occurrence for each category. This are all the
paths that we can have in the system:
Frequency computation for each type of product:

Now we can compute the different Ws of the different categories of customer (lead
time or throughput time).

-Categorization centre (that has priority rule):

-Ophthalmology (that has priority rule):

-Orthopaedics:

(Lq computed with the table, with interpolation; at the exam isn’t necessary
interpolation)

-General surgery:

-Gynaecology:

-OSS:
In the preparing phase we have two systems, each having its own queue; the arrival
rate is the overall rate divided by two, because each system is going to see one single
queue. Lambda = 15/2.

 Mean Ws for the whole system:

By summing the different Ws in each stage (for instance nurse, ortop. and OSS) and
multiplying for the occurrences we obtain the weighted time.
The sum of the values in the last column gives the total Ws.
EXERCISE 2.3

Note: “not considering scraps paths” or “consider all the customers that enter all the
stages” are the same thing (useful for the exam).

Using different priorities, the overall throughput time of the system will be the same.

WIP before each stage = Lq

SOL:

We start to compute all the lambda.

(Node balancing must be considered for the computation, what enters must be exactly
the same of what exit).

Resuming the information that we have, now we can compute the Ws (or LT, lead
time):
Now we have to map our system because we are going to have different types of
customers.
We have 7 different types.

The text asked to look at those paths that are not discarded but that exit the system.
So, we will not consider the paths A1, A6 and A7.
Anyway, we need to compute the occurrences of all the path to have information
about the 4 occurrences once eliminated the others.

-A1 occurrence:
Note: the occurrence computation of a certain population refers to the input, so we
need to divide by the input of the system.

-A2, A3, A6 occurrences:

Population that doesn’t go to the assembly 1 twice.

-A4, A5, A7 occurrences:


Adjusted occurrences:

We need to compute the adjusted occurrences in order to compute the Ws.

Throughput time
EXERCISE 3.1 23/03
NOTE for all the exercises:
Steps for the computation of Ws (throughput time) of the system:
1. Calculate the Ws of each stage
2. Understand the different paths
3. Calculate the occurrences
4. Calculate the Ws of the system

SOL:

INPUT A:

We have information about the output (300 products must be delivered).

Node balancing:

Nota: l’analisi dei nodi solitamente parte dal nodo più lontano fino a quello più vicino!!
(credo)
Nodes must be individuated in the best option you think.

INPUT B:

Node balancing:

Lambda (arrival rates) computed for each stage:

Now we start computing the Ws of each stage:
Computation of the occurrences of each path:
A)

B)

- In B1, B3 and B4 paths the pieces haven’t to be reworked by finishing and drilling
machines. The total amount of these pieces is 0,8*XB. 0,8*XB is distributed
among good pieces at first test, good pieces at 2°test and scraps at second test.
The quantities of pieces that enter in the testing phase are 0,8*XB that arrives
directly from work center and 0,15*XB is reworked by traditional machines This
proportion is kept also for the pieces processed by the testing machine and the pieces
that have to be tested again.
0,8/(0,8+0,15) = 0,8/0,95 not-reworked
0,15/(0,8+0,15) = 0,15/0,95 reworked

- In B5, B6 and B7 paths the pieces have to be reworked by finishing and drilling
machines.
The total amount of these pieces is 0,15*XB. 0,15*XB is distributed among good
pieces at first test, good pieces at 2°test and scraps at second test

Now we need to compute the adjusted occurrences, to eliminate from the

computations those paths that are going to be discarded. As numerator we take the
occurrences just computed. The denominator is the sum of all those paths that are
not discarded.
Adjusted occurrences must be computed every time is requested to compute the Ws
without considering the paths that are discarded.
Now we can compute for every path the weighted time.
EXERCISE 3.2
Q1: Map of the system (and computation of the input):

Q2: WIP and Wq (quality control stage):

Q3: Probability of Ws lower than 45 sec.

(non ho capito questo punto, rivedi)
Q4: average throughput time

Ws mechanical process: