1.

225J
1.225J (ESD 225) Transportation Flow Systems

Lecture 8

Delays in Probabilistic Models:

Elements from Queueing Theory

Profs. Ismail Chabini and Amedeo Odoni

Lecture 8 Outline

‰ Introduction to Queueing
‰ Conceptual Representation of Queueing Systems
‰ Codes for Queueing Models
‰ Terminology and Notation
‰ Little’s Law and Basic Relationships
‰ Exponential Distribution for Interarrival and Service times Modeling
‰ State Transition Diagram
‰ Derivation of waiting characteristics for M/M/1
‰ Summary

1.225, 11/26/02 Lecture 8, Page 2

 Applications of Queueing Theory

‰ Some familiar queues:
• Airport check-in
• Automated Teller Machines (ATMs)
• Fast food restaurants
• On hold on an 800 phone line
• Urban intersection
• Toll booths
• Aircraft in a holding pattern
• Calls to the police or to utility companies
‰ Level-of-service (LOS) standards
‰ Economic analyses involving trade-offs among operating costs, capital
investments and LOS
‰ Queueing theory predicts various characteristics of waiting lines (or
queues) such as average waiting time

1.225, 11/26/02 Lecture 8, Page 3

Queueing Models Can Be Essential in Analysis of Capital

Investments

Cost
Total cost

Optimal
cost Cost of building the capacity

Cost of losses due to waiting

“Optimal” capacity Capacity

1.225, 11/26/02 Lecture 8, Page 4

2
 Strengths and Weaknesses of Queueing Theory

‰ Queueing models necessarily involve approximations and

simplification of reality
‰ Results give a sense of order of magnitude, of changes relative to
a baseline, of promising directions in which to move
‰ Closed-form results are essentially limited to “steady state”
conditions and derived primarily (but not solely) for birth-and-
death systems and “phase” systems
‰ Some useful bounds for more general systems at steady state
‰ Numerical solutions are increasingly viable for dynamic systems

1.225, 11/26/02 Lecture 8, Page 5

Queueing Process and Queueing System

Queueing System
Servers

C
Arrival point Departure point
at the system C from the system

Queue C
Source
of users/ C C C C C C C
customers
C
C
C

Arrivals
process
Size of Queue discipline and Service process Number of servers
user source Queue capacity
1.225, 11/26/02 Lecture 8, Page 6

 Queueing network consisting of five queueing systems

Queueing Queueing
system system
2 3

Queueing Point where Point where Queueing Out

In
system users make users merge + system
1 a choice 5

Queueing
system
4

1.225, 11/26/02 Lecture 8, Page 7

A Code for Queueing Models: A/B/m

Distribution of Queueing System

service time
Number of servers
–/–/– C S
Queue
Customers C S Service
Distribution of CCCCCC
C S facility
interarrival time
C S

‰ Some standard code letters for A and B:

• M: Negative exponential (M stands for memoryless)
• D: Deterministic
• Ek:kth-order Erlang distribution
• G: General distribution
‰ Model covered in this lecture: M/M/1
1.225, 11/26/02 Lecture 8, Page 8

 Terminology and Notation

‰ State of system: number of customers in queueing system
‰ Queue length: number of customers waiting for service
‰ N(t) = number of customers in queueing system at time t
‰ Pn(t) = probability that N(t) is equal to n
‰ λn: mean arrival rate of new customer when N(t) = n
‰ µn: mean (combined) service rate when N(t) = n
‰ Transient condition: state of system at t depends on the state of the
system at t=0 or on t
‰ Steady state condition: system is independent of initial state and t
‰ s: number of servers (parallel service channels)
‰ If λn and the service rate per busy server are constant, then λn=λ, µn=sµ
1
‰ Expected interarrival time =
λ
1
‰ Expected service time = µ
1.225, 11/26/02 Lecture 8, Page 9

Quantities of Interest at Steady State

‰ Given:
• λ = arrival rate
• µ = service rate per service channel (number of servers =1, in this
lecture)
‰ Unknowns:
• L = expected number of users in queueing system
• Lq = expected number of users in queue
• W = expected time in queueing system per user (W = E(w))
• Wq = expected waiting time in queue per user (Wq = E(wq))

‰ 4 unknowns ⇒ We need 4 equations

1.225, 11/26/02 Lecture 8, Page 10

 Little’
Little’s Law

Number of A(t): cumulative arrivals to the system

users C(t): cumulative service completions in the system

A(t)

N(t)

C(t)

t T Time
T T
∫
LT = 0
N (t )dt
=
A(T ) ∫0 N (t ) dt
⋅ = λT ⋅ WT
1.225, 11/26/02
T T A(T ) Lecture 8, Page 11

Relationships between L, Lq, W,

W, and Wq
‰ 4 unknowns: L, W, Lq, Wq
‰ Need 4 equations. We have the following 3 equations:
• L = λW (Little’s law)
• Lq = λWq
1
• W = Wq +
µ
‰ If we know L (or any one of the four expected values), we can determine
the value of the other three
‰ The determination of L may be hard or easy depending on the type of
queueing model at hand (i.e. M/M/1, M/M/s, etc.)
∞
‰ L = ∑ nPn ( Pn : probability that n customers are in the system)
n=0

1.225, 11/26/02 Lecture 8, Page 12

 Modeling Interarrival Time and Service Time

• T : Interarrival (service) time random variable

αe −αt , t ≥ 0
• Density function : fT (t) = 

0 , t < 0

1 1
• P{0 ≤ T ≤ t} = 1− e −αt , E (T ) = , var(T ) =
α α2

• For small ∆t, P{0 ≤ T ≤ ∆t} ≈ α ∆t (why?)

∞
xk
• e x = 1+ x + ∑
k =2 k!
∞
(−α ∆t) k
• P{0 ≤ T ≤ ∆t} = 1− e −α ∆t = 1− (1− α ∆t + ∑
)
k =2 k!

≈ α ∆t (for small ∆t)

• Interarrival Time : α = λ ; Service Time : α = µ

1.225, 11/26/02 Lecture 8, Page 13

State Transition Diagram for M/M/1

‰ States:
0 1 2 … n-1 n n+1

‰ During ∆t:

P0 λ∆t P1λ∆t P2 λ∆t Pn − 2 λ∆t Pn −1λ∆t Pn λ∆t Pn +1λ∆t

0 1 2 … n-1 n n+1

P1µ∆t P2 µ∆t P3 µ∆t Pn −1µ∆t Pn µ ∆t Pn +1µ∆t Pn + 2 µ∆t

‰ Another way to represent it: State Transition Diagram

λ λ λ λ λ λ λ
0 1 2 … n-1 n n+1

µ µ µ µ µ µ µ

1.225, 11/26/02 Lecture 8, Page 14

7
 Observing State Transition Diagram from Two Points
‰ From point 1:

λP0 = µP1 (λ + µ ) P1 = λP0 + µP2 (λ + µ ) Pn = λPn −1 + µPn +1

λ λ λ λ λ λ λ
0 1 2 … n-1 n n+1

µ µ µ µ µ µ µ

‰ From point 2:

λP0 = µP1 λP1 = µP2 λPn = µPn +1

λ λ λ λ λ λ λ
0 1 2 … n-1 n n+1

µ µ µ µ µ µ µ

1.225, 11/26/02 Lecture 8, Page 15

Derivation of P0 and Pn
2 n
λ λ λ
‰ Putting it all together: P1 = P0 , P2 =   P0 , L, Pn =   P0
µ µ µ
n
∞ ∞
λ 1
‰ Since ∑ Pn = 1, ⇒ P0 ∑   = 1 ⇒ P0 =
n =0  µ 
n
∞
λ
∑
n =0
 
n =0  µ 
n
λ ∞
λ ∞
1− ρ ∞ 1
‰ Let ρ =
µ
, then ∑   = ∑ ρ n =
n= 0  µ − ρ
=
− ρ
(Q ρ < 1)
 n =0 1 1

1
‰ Therefore, P0 = ∞
= 1− ρ and Pn = ρ n (1 − ρ )
∑ρ
n =0
n

1.225, 11/26/02 Lecture 8, Page 16

 Derivation of L, W, Wq, and Lq

∞ ∞ ∞ ∞
• L = ∑ nPn =∑ nρ n (1 − ρ ) = (1 − ρ )∑ nρ n = (1 − ρ ) ρ ∑ nρ n −1
n =0 n =0 n =0 n =1

d  ∞ n d  1 
= (1 − ρ ) ρ  ∑ ρ  = (1 − ρ ) ρ  
dρ  n = 0  dρ  1 − ρ 
λ
 1  ρ µ λ
= (1 − ρ ) ρ  =
2 
= =
λ
 (1 − ρ )  (1 − ρ ) 1 − µ µ − λ
L λ 1 1
•W= = ⋅ =
λ µ −λ λ µ −λ

1 1 1 λ
• Wq = W − = − =
µ µ −λ µ µ (µ − λ )

λ λ2
• Lq = λ Wq = λ ⋅ =
µ (µ − λ ) µ (µ − λ )

1.225, 11/26/02 Lecture 8, Page 17

Lecture 8 Summary

‰ Introduction to Queueing
‰ Conceptual Representation of Queueing Systems
‰ Codes for Queueing Models
‰ Terminology and Notation
‰ Little’s Law and Basic Relationships
‰ Exponential Distribution for Interarrival and Service times Modeling
‰ State Transition Diagram
‰ Derivation of waiting characteristics for M/M/1

1.225, 11/26/02 Lecture 8, Page 18