Course: Operations Research (FIN 309)
Queuing Theory
Shahadat Hossain, Ph.D.
Department of Finance
University of Chittagong
April 2026
1 / 42
Outline
1 Introduction & Key Terminology
2 Single Channel Model (M/M/1)
3 Multiple Channel Model (M/M/k)
4 Solved Examples
5 Summary & Comparison
2 / 42
What is Queuing Theory?
Definition
Queuing Theory is the mathematical study of waiting lines (queues) — used to predict queue
lengths, waiting times, and system efficiency.
Queuing System
A system consisting of:
– Arrival of customers
– Waiting in a queue
– Being picked up for service (according to a discipline)
– Being serviced and departing
3 / 42
Key Terms — Part I
Basic Concepts
– Customer: Person, machine, or unit
arriving for service
System Measures
– Service Station: Point where service is – Time in System: Waiting time + Service
provided time
– Waiting Time: Time spent in the queue – Customers in System: Customers in
before service queue + customers being served
– Queue Length: Number of customers
waiting in the queue
4 / 42
Key Terms — Part II: Customer Behavior
Three Types of Customer Behavior
1 Balking
Customer decides not to join the queue at all
2 Reneging
Customer joins the queue but leaves it afterwards
3 Jockeying
Customer switches from one queue to another
5 / 42
Service / Queue Disciplines
Common Queue Disciplines
Discipline Description
FIFO / FCFS First In, First Out / First Come, First Served
LIFO / LCFS Last In, First Out / Last Come, First Served
SIRO Service In Random Order
Priority Customers served based on assigned priority
Most Common
FIFO/FCFS is the most commonly used discipline in practice.
6 / 42
Single Channel Model — Overview
Assumptions
– Arrivals follow a Poisson process with rate λ
– Service times follow an Exponential distribution with rate µ
– Single server, infinite queue capacity
– System is in steady state: λ < µ
Traffic Intensity
λ
ρ= <1 (required for stability)
µ
7 / 42
Single Channel Model — Formulas (M/M/1)
Measure Formula
λ
Prob. no units in system P0 = 1 −
µ
λ2
Avg. units in queue Lq =
µ(µ − λ)
λ λ
Avg. units in system L = Lq + =
µ µ−λ
Lq λ
Avg. time in queue Wq = =
λ µ(µ − λ)
1 1
Avg. time in system W = Wq + =
µ µ−λ
λ
Prob. unit must wait Pw =
µ
8 / 42
Single Channel Model — Probability of n Units
Probability of exactly n units in the system
n
λ
Pn = P0
µ
Useful Tail Probabilities
k+1
λ
P (n > k) =
µ
k k+1
λ λ λ
P (n ≥ k) = 1− +
µ µ µ
k
λ λ
P (n = k) = 1−
µ µ
9 / 42
Multiple Channel Model — Overview
Assumptions
– k servers, each with service rate µ
– Arrivals follow Poisson process with rate λ
– Service times exponentially distributed
– Steady-state condition: λ < kµ
Traffic Intensity
λ
ρ= <1
kµ
10 / 42
Multiple Channel Model — Key Formulas (M/M/k)
Probability of Zero Units in System
1
P0 = k−1
X (λ/µ)n (λ/µ)k kµ
+ ·
n! k! kµ − λ
n=0
Average Number in Queue
(λ/µ)k λµ
Lq = · P0
(k − 1)!(kµ − λ)2
11 / 42
Multiple Channel Model — Derived Measures
Measure Formula
λ
Avg. units in system L = Lq +
µ
Lq
Avg. time in queue Wq =
λ
1
Avg. time in system W = Wq +
µ
k
1 λ kµ
Prob. unit must wait Pw = P0
k! µ kµ − λ
12 / 42
Multiple Channel — Probability of n Units
Probability of n units in the system
(λ/µ)n
n! P0
for n ≤ k
Pn =
(λ/µ)n
P0 for n > k
k! k n−k
13 / 42
Example 1 — Petrol Pump: Question
Problem
At a certain petrol pump, customers arrive in a Poisson process with an average time of 5
minutes between arrivals. The time-intervals between services follow an exponential
distribution with a mean service time of 2 minutes. Answer the following:
1 What is the expected average queue length?
2 What is the average number of customers in the queuing system?
3 How long on average does a customer wait in the queue?
4 How much time on average does a customer spend in the system?
5 By how much should the flow of customers be increased to justify opening a second
service point, if management is willing to do so only when a customer has to wait 5
minutes for service?
14 / 42
Example 1 — Petrol Pump: Solution (i–iv)
Rate Derivation
Inter-arrival time = 5 min ⇒ λ = 15 × 60 = 12 customers/hr Service time = 2 min
⇒ µ = 21 × 60 = 30 customers/hr Traffic intensity: ρ = 3012
= 0.4 (stable system)
Measure Formula Result
λ2 144
(i) Avg. queue length Lq = 0.267 customers
µ(µ − λ) 30 × 18
λ 12
(ii) Avg. in system L = 0.667 customers
µ−λ 18
λ 12
(iii) Avg. wait in queue Wq = hr 1.33 minutes
µ(µ − λ) 540
1 1
(iv) Avg. time in system W = hr 3.33 minutes
µ−λ 18
15 / 42
Example 1 — Petrol Pump: Solution (v)
When is a Second Service Point Justified?
1
Management opens a second point only if Wq ≥ 5 min = hr.
12
Current Wq = 1.33 min, so the threshold is not yet reached.
1
We need to find the new arrival rate λ1 at which Wq = hr:
12
Solving for λ1
λ1 1
Wq = = =⇒ 12λ1 = µ(µ − λ1 ) = 30(30 − λ1 )
µ(µ − λ1 ) 12
12λ1 = 900 − 30λ1 =⇒ 42λ1 = 900 =⇒ λ1 ≈ 21.43 customers/hr
Interpretation
The arrival rate must rise from 12 to 21.43 per hour — an increase of 78.6% — before a second
service point is warranted.
16 / 42
Example 2 — Jewellery Store Salesman: Question
Problem
Amin Jewellery Store seeks a salesman for its evening shift. Three applicants have applied,
each demanding different salaries:
Applicant Hourly Wage Avg. Service Time
Mr. Karim $6 6 minutes
Mr. Rahim $10 5 minutes
Mr. Selim $14 4 minutes
Customers arrive at 8 per hour. Cost of a customer in the system = $4/hr per customer.
Which applicant should be hired?
17 / 42
Example 2 — Jewellery Store Salesman: Solution
λ
Method: Total Hourly Cost = Wage + $4 × L, where L = , λ = 8/hr
µ−λ
8
Applicant Wage Svc time µ (hr) L= µ−8
Total Cost
Mr. Karim $6 6 min 60
6 = 10 8
2 = 4.00 6 + 4(4.00) = $22.00
Mr. Rahim $10 5 min 60
5 = 12 8
4 = 2.00 10 + 4(2.00) = $18.00
Mr. Selim $14 4 min 60
4 = 15 8
7 = 1.14 14 + 4(1.14) = $18.56
Decision
Although Mr. Selim serves fastest (µ = 15/hr), his higher wage makes him marginally costlier than Mr.
Rahim. Mr. Rahim minimises total cost at $18.00/hr.
18 / 42
Example 3 — Repairman Selection: Question
Problem
Machines break down at 5 per hour (Poisson). Idle machine cost = $15/hr. Two repairmen
interviewed:
Repairman Hourly Charge Service Rate (µ)
A $8/hr 7 machines/hr
B $10/hr 9 machines/hr
Work shift = 8 hours. Which repairman should be used and why?
19 / 42
Example 3 — Repairman Selection: Solution
Key Idea
λ
L= gives the average number of machines in the system. Since each machine in the system is
µ−λ
idle, L machine-hours are lost per hour. Over an 8-hour shift: machine-hours lost = L × 8.
Repairman A (µ = 7, $8/hr) Repairman B (µ = 9, $10/hr)
5 5
L= 7−5 = 2.5 machines L= 9−5 = 1.25 machines
Idle machine-hrs = 2.5 × 8 = 20 Idle machine-hrs = 1.25 × 8 = 10
Idle cost = 20 × $15 = $300 Idle cost = 10 × $15 = $150
Hire cost = 8 × $8 = $64 Hire cost = 8 × $10 = $80
Total = $300 + $64 = $364 Total = $150 + $80 = $230
Decision
Repairman B should be hired. Despite a higher hourly rate, the faster service reduces idle machine costs
enough to save $134 per shift.
20 / 42
Example 4 — Bank Tellers: Question
Problem
A bank has two tellers working on savings accounts. The first teller handles withdrawals only ;
the second handles deposits only. Service times for both are exponential with a mean of 3
minutes per customer.
– Depositors arrive in a Poisson fashion at 16 per hour
– Withdrawers arrive in a Poisson fashion at 14 per hour
What would be the effect on the average waiting time for depositors and withdrawers
if each teller handles both withdrawals and deposits?
21 / 42
Example 4 — Bank Tellers: Solution
60
Rate Derivation: Mean service time = 3 min ⇒ µ = 3
= 20/hr for each teller
Case 1: Separate Queues Case 2: Combined (k = 2, λ = 30, µ = 20)
Depositors (λ1 = 16, µ = 20):
16 16 1 1
Wq,1 = = = 0.2 hr = 12 min P0 = P 1 (1.5)n (1.5)2
=
20(20 − 16) 80 + · 40 7
n=0 n! 2! 10
Withdrawers (λ2 = 14, µ = 20): (1.5)2 · 20 1 45
Lq = · =
1! · (10)2 7 700
14 14
Wq,2 = = = 0.117 hr = 7 min Lq 45
20(20 − 14) 120 Wq = = ≈ 0.064 hr = 3.86 min
λ 700 × 30
Conclusion
Combining reduces depositor wait from 12 min and withdrawer wait from 7 min to a single shared wait of
3.86 min — a significant improvement for both groups.
22 / 42
Example 5 — Post Office Counters: Question
Problem
A post office has two counters. The first counter handles money orders and registered letters;
the second handles all other business. Service times for both are exponential with a mean of
4 minutes per customer.
– Customers arrive at the first counter at 10 per hour (Poisson)
– Customers arrive at the second counter at 12 per hour (Poisson)
What would be the effect on the average waiting time for both types of customers if
each counter can handle all types of business?
23 / 42
Example 5 — Post Office Counters: Solution
60
Rate Derivation: Mean service time = 4 min ⇒ µ = 4
= 15/hr per counter
Case 1: Separate Counters Case 2: Combined (k = 2, λ = 22, µ = 15)
Counter 1 (λ1 = 10, µ = 15):
10 10 1
Wq,1 = = = 0.133 hr = 8 min P0 = P1 (22/15)n (22/15)2
15(5) 75 30
n=0 n! + 2! · 8
Counter 2 (λ2 = 12, µ = 15): = 0.1538
12 12 (22/15)2 · 15
Wq,2 = = = 0.267 hr = 16 min Lq = · 0.1538
15(3) 45 1! · (8)2
Lq
Wq = = 0.0775 hr = 4.65 min
22
Conclusion
The busier counter 2 had customers waiting up to 16 min. Merging the queues reduces all
customers to 4.65 min — a 71% reduction.
24 / 42
Example 6 — Standard M/M/1: Question
Problem
Given the following information:
– Arrivals and services follow a Poisson process
– Customers arrive at a rate of 8 per hour
– Service rate is 10 customers per hour
Answer the following:
1 What is the average number of customers waiting for service?
2 What is the average time a customer must wait in the queue?
3 What is the average time for a customer to be in the system?
25 / 42
Example 6 — Standard M/M/1: Solution
Given: λ = 8/hr, µ = 10/hr ⇒ Traffic intensity: ρ = λ/µ = 0.8 (system is stable since
ρ < 1)
Calculations Interpretation
– On average 3.2 customers are waiting in the
queue at any time
P0 = 1 − ρ = 1 − 0.8 = 0.2 (20% idle time)
– A customer spends 24 minutes waiting
λ2 64 before being served
Lq = = = 3.2 customers
µ(µ − λ) 10 × 2 – Total time from arrival to departure is 30
minutes
Lq 3.2
Wq = = = 0.4 hr (24 min) – The server is busy 80% of the time (ρ = 0.8)
λ 8
1 1
W = = = 0.5 hr (30 min)
µ−λ 2
26 / 42
Example 7 — Hospital Cost & Probability: Question
Problem
Given: Mean arrival rate λ = 4 patients/hr, mean service rate µ = 5 patients/hr. The facility
operates 24 hours a day. Cost for each hour lost due to waiting = Tk. 10. Arrivals and
services follow a Poisson process.
Find:
1 Average cost per day from waiting
2 Impact of increasing µ to 6/hr on average daily cost
3 Savings in daily cost when µ becomes 6/hr
4 P (n > 2) and P (n > 4)
5 P (n ≥ 4)
6 Probability that the queue length is exactly 3
7 P (n ≤ 4)
27 / 42
Example 7 — Hospital: Solution (i–iii) Cost Analysis
Given: λ = 4/hr, µ = 5/hr, 24-hr day, 96 arrivals/day, cost = Tk. 10/hr/patient
(i) Current daily cost (µ = 5) (ii) Improved daily cost (µ = 6)
λ 4 4
Wq = = = 0.8 hr Wq = = 13 hr
µ(µ − λ) 5×1 6×2
Total wait/day = 96 × 0.8 = 76.8 hr Total wait/day = 96 × 31 = 32 hr
Daily cost = 76.8 × 10 = Tk. 768 Daily cost = 32 × 10 = Tk. 320
(iii) Savings in system time
With µ = 5: W = 11 = 1 hr/patient ⇒ system loss = 96 × 1 = 96 hr = Tk. 960
With µ = 6: W = 12 = 0.5 hr/patient ⇒ system loss = 96 × 0.5 = 48 hr = Tk. 480
Daily savings = Tk. 960 − Tk. 480 = Tk. 480
28 / 42
Example 7 — Hospital: Solution (iv–vii) Probabilities
Given: ρ = λ/µ = 4/5 = 0.8, using P (n > k) = ρk+1 , P (n = k) = ρk (1 − ρ)
Part Expression Result
(iv-a) P (n > 2) (0.8)2+1 = (0.8)3 0.512
(iv-b) P (n > 4) (0.8)4+1 = (0.8)5 0.328
(v) P (n ≥ 4) (0.8)4 (1 − 0.8) + (0.8)5 = 0.08192 + 0.32768 0.4096
(vi) P (n = 4) Queue length = 3 means system has 4 patients
P (n = 4) = (0.8)4 (1 − 0.8) 0.08192
(vii) P (n ≤ 4) 1 − P (n > 4) = 1 − 0.32768 0.67232
Note on Part (vi)
Queue length = 3 exactly means 3 patients are waiting and 1 is being served, so the total number in the
system = 4.
29 / 42
Example 8 — Railway Station Reservation: Question
Problem
In the central railway station, 15 computerised reservation counters are available. A customer can
book a ticket for any train on any day at any one of these counters. The average time spent per
customer by each clerk is 5 minutes (µ = 12/hr per counter). Average arrivals per hour and the
maximum willing wait time per period are:
Period Total Arrivals (λ′ ) Max. Willing Wait
(i) Peak 110 per hour 15 minutes
(ii) Normal 60 per hour 10 minutes
(iii) Low 30 per hour 5 minutes
Making suitable assumptions, determine how many counters should be kept open during
each type of period.
30 / 42
Example 8 — Railway Station: Solution
λ
Method: µ = 12/hr per counter. Set Wq = equal to the max. wait, solve for
µ(µ − λ)
λ per counter, then k = ⌈λ′ /λ⌉.
Period λ′ , max Wq Equation & solution Counters
15 λ
(i) Peak 110/hr, 15 min 60 = 12(12−λ) ⇒ λ = 9/hr ⌈110/9⌉ = 13
10 λ
(ii) Normal 60/hr, 10 min 60 = 12(12−λ) ⇒ λ = 8/hr ⌈60/8⌉ = 8
5 λ
(iii) Low 30/hr, 5 min 60 = 12(12−λ) ⇒ λ = 6/hr ⌈30/6⌉ = 5
Conclusion
Of the 15 available counters: 13 open during peak, 8 during normal, and 5 during low-activity periods.
31 / 42
Example 9 — Tool Crib : Question
Problem
Arrivals of machinists at a tool crib are considered to be Poisson distributed at an average
rate of 6 per hour. The length of time a machinist must remain at the tool crib is
exponentially distributed with the average time being 0.05 hours (i.e. 3 minutes).
Answer the following:
(i) What is the probability that a machinist arriving at the tool crib will have to wait?
(ii) What is the average number of machinists at the tool crib?
(iii) The company will install a second tool crib when convinced that a machinist would
expect to spend at least 6 minutes waiting and being serviced at the tool crib. By how
much should the flow of machinists increase to justify the addition of a second tool crib?
32 / 42
Example 9 — Tool Crib: Solution
Rate Derivation: λ = 6/hr, Mean service time = 0.05 hr ⇒ µ = 1/0.05 = 20/hr,
ρ = 6/20 = 0.3
(i) Probability of Waiting (iii) Threshold for Second Tool Crib
Second crib justified when W ≥ 6 min = 0.1 hr.
λ 6
Pw = = = 0.3 1
µ 20 W = = 0.1 ⇒ µ − λ1 = 10
µ − λ1
There is a 30% chance an arriving machinist
must wait. λ1 = 20 − 10 = 10 per hour
Current flow = 6/hr. Required flow = 10/hr.
(ii) Average Number at Tool Crib Flow must increase by 4 machinists/hr (a 67%
increase).
λ 6 6
L= = = ≈ 0.43 machinists
µ−λ 20 − 6 14
33 / 42
Example 10 — Car Service Station : Question
Problem
Arrival rate of cars to a certain service station follows a Poisson distribution with an average
time of 50 minutes between two consecutive arrivals. The service time is exponentially
distributed with a mean of 25 minutes.
Determine the following:
(a) The probability that a car arriving at the station will have to wait.
(b) The probability that a car will have to wait for more than 10 minutes.
(c) The probability that a car will have to wait for more than 20 minutes.
(d) Find the flow rate of cars if the average waiting time in the queue is 35 minutes.
(e) What is the utilisation fraction of the service station?
34 / 42
Example 10 — Car Service Station: Solution
Rate Derivation: Inter-arrival = 50 min ⇒ λ = 1.2/hr; Service = 25 min ⇒ µ = 2.4/hr;
ρ = 0.5
(a) Probability of Waiting (d) Flow Rate when Wq = 35 min
λ 1.2 λ 35
Pw = ρ = = = 0.5 Wq = = hr
µ 2.4 µ(µ − λ) 60
Half of all arriving cars must wait. 60λ = 35 × 2.4(2.4 − λ)
60λ = 84(2.4 − λ) = 201.6 − 84λ
−(µ−λ)t
(b) & (c) P (Wq > t) = ρ · e 144λ = 201.6 ⇒ λ = 1.4 cars/hr
µ − λ = 1.2/hr = 0.02/min
P (Wq > 10) = 0.5 e−0.02×10 = 0.5 e−0.2 ≈ 0.409 (e) Utilisation Fraction
−0.02×20 −0.4
P (Wq > 20) = 0.5 e = 0.5 e ≈ 0.335
λ 1.2 35 / 42
Example 11 — Telephone Booth : Question
Problem
At a telephone booth, arrivals are considered to be Poisson with an average inter-arrival time
of 12 minutes. The length of the phone call may be assumed to be exponentially
distributed with an average of 4 minutes.
Calculate the following:
(i) The probability that a fresh arrival will not have to wait for the phone.
(ii) The probability that an arrival will have to wait for more than 10 minutes before the
phone is free. [optional]
(iii) The average length of queues formed from time to time.
36 / 42
Example 11 — Telephone Booth: Solution
1 λ 1/12 1
Rate Derivation: λ = 12
/min, µ = 14 /min, ρ= = = ≈ 0.33
µ 1/4 3
(i) P(fresh arrival need not wait) (ii) P(wait > 10 min)
Using the waiting-time distribution:
P0 = 1 − ρ = 1 − 0.33 = 0.67 Z ∞
λ −(µ−λ)s
67% of arrivals find the booth free and proceed P (Wq > t) = µ (µ − λ) e ds
t
immediately. Z ∞
P (Wq > 10) = (0.33)(0.167) e−0.167t dt
10
(iii) Avg. queue length (given L > 0)
= 0.0621
µ 0.25 0.25
L= = = = 1.5
µ−λ 0.25 − 0.083 0.167
When a queue exists, it averages 1.5 callers.
37 / 42
Example 12— Petrol Station, Two Pumps : Question
Problem
A petrol station has two pumps. The service time follows an exponential distribution with a
mean of 4 minutes. Cars arrive for service in a Poisson process at the rate of 10 cars per
hour.
(i) Find the probability that a customer has to wait for service.
(ii) For what proportion of time does the pump remain idle?
38 / 42
Example 12 — Petrol Station, Two Pumps: Solution
60 λ 10 1
Rate Derivation: s = 2, λ = 10/hr, µ= 4
= 15/hr, ρ= = =
sµ 2 × 15 3
Computing P0 (i) P(customer must wait)
#−1
(λ/µ)s (2/3)2 1 1
" s−1
X (sρ)n (sρ)s
P0 = + Ps = P0 = × =
n! s!(1 − ρ) s! 2! 2 9
n=0
Ps 1/9 1
(1/3×2)2
−1 P (w > 0) = = = ≈ 0.167
= 1 + 31 ×2 + 1−ρ 2/3 6
2!(1 − 1/3)
−1
= 1 + 32 + 94 × 12 × 32
(ii) Pump idle proportion
2 1 −1
1
= 1+ 3 + 3 = Pump utilisation = ρ = 31 .
2
1 2
Proportion idle = 1 − ρ = 1 − 3 = 3 ≈ 66.7%
39 / 42
Single vs. Multiple Channel — Summary
M/M/1 (Single) M/M/k (Multiple)
Servers 1 k
Stability λ<µ λ < kµ
P0 1 − λ/µ Complex formula
λ2
Lq µ(µ−λ) Uses P0
1 1
W µ−λ Wq + µ
Best for Low traffic High traffic
40 / 42
Key Insights & Takeaways
Practical Insights
1 Waiting time grows non-linearly as λ → µ — small increases in load cause large
increases in wait
2 Combining queues under multiple servers almost always reduces average waiting time
3 Optimal staffing balances wage costs against customer waiting costs
4 Higher service rate can justify significant cost savings in the long run
Golden Rule
Total Cost = Service Cost + Waiting Cost = Wages + c · L
41 / 42
Thank You
Questions & Discussion