1 Queueing Analysis of Production Systems (Factory Physics)

Preview:

Citation preview

1

Queueing Analysis of Production Systems

(Factory Physics)

2

Reading Material

Chapter 8 from textbook

Handout: Single Server Queueing Model by Wallace Hopp (available for download from class website)

3

Queueing analysis is a tool for

evaluating operational performance Utilization

Time-in-system (flow time, leadtime)

Throughput rate (production rate, output

rate)

Waiting time (queueing time)

Work-in-process (number of parts or batches

in the systems)

4

A Single Stage System

Raw material

Processing unit

Finished parts

5

The Queueing Perspective

Server (production facility)

Queue (logical or physical) of

jobs

Arrival (release) of

jobs

Departure (completion) of

jobs

6

E[A]: average inter-arrival time between

consecutive jobs : arrival rate (average number of jobs that arrive per unit

time), = 1/E[A] E[S]: average processing time : processing rate (maximum average number of jobs that can be processed per unit time), = 1/E[S]

: average utilization, = E[S]/E[A] = /

System Parameters

7

E[W]: average time a job spends in the

system

E[Wq]: average time a job spends in the

queue

E[N]: average number of job in the system (average WIP in the system)

E[Nq]: average number of jobs in the queue (average WIP in the queue)

TH: throughput rate (average number of jobs produced per unit time)

Performance Measures

8

E[W] = E[Wq] + E[S]

E[N] = E[Nq] +

Performance Measures (Continued…)

9

Little’s Law

10

E[N] = E[W]E[Nq] = E[Wq]

= E[S]

Little’s Law

11

Example 1

Jobs arrive at regular & constant intervals

Processing times are constant Arrival rate < processing rate ( < )

12

Example 1

Jobs arrive at regular & constant intervals Processing times are constant Arrival rate < processing rate ( < )

E[Wq] = 0

E[W] = E[Wq] + E[S] = E[S] = / E[N] = E[W] = E[S] = E[Nq] = E[Wq] = 0 TH =

13

Case 2

Jobs arrive at regular & constant intervals

Processing times are constant Arrival rate > processing rate ( > )

14

Case 2

Jobs arrive at regular & constant intervals Processing times are constant Arrival rate > processing rate ( > )

E[Wq] = E[W] = E[Wq] + E[S] = = 1E[N] = E[W] = E[Na] = E[Wa] = TH =

15

Case 3

Job arrivals are subject to variability Processing times are subject to variability Arrival rate < processing rate ( < )

Example:

Average processing time = 6 min Inter-arrival time = 8 min

0.25y probabilitmin with 2

0.50y probabilitmin with 6

0.25y probabilitmin with 10

timeProcessing

16

Case 3 (Continued…)

= 6/8 = 0.75 TH = 1/8 job/min = 7.5 job/hour E[Wq] > 0 E[W] > E[S] E[Nq] > 0 E[N] >

17

In the presence of variability, jobs may wait for processing and a queue in front of the processing unit may build up.

Jobs should not be released to the system at a faster rate than the system processing rate.

18

Sources of Variability

19

Sources of Variability

Sources of variability include: Demand variability Processing time variability Batching Setup times Failures and breakdowns Material shortages Rework

20

Measuring Variability

Var[ ] Variance of processing time

Var[ ] Variance of inter-arrival time

Var[ ]Coefficient of variation (CV) in processing time

[ ]

Var[A]Coefficient of variation (CV) in inter-arrival time

[ ]

S

A

S

A

SC

E S

CE A

21

Variability Classes

0.75

High variability(HV)

Moderate variability(MV)

Low variability(LV)

0 1.33CV

22

Illustrating Processing Time VariabilityTrial Machine 1 Machine 2 Machine 3

1 22 5 5 2 25 6 6 3 23 5 5 4 26 35 35 5 24 7 7 6 28 45 45 7 21 6 6 8 30 6 6 9 24 5 5

10 28 4 4 11 27 7 7 12 25 50 500 13 24 6 6 14 23 6 6 15 22 5 5

Mean 25.1 13.2 43.2 Std dev 2.5 15.9 127.0

CV 0.1 1.2 2.9 CV 0.01 1.4 8.6

Class LV MV HV

23

Illustrating Arrival Variability

t

Low variability arrivals

t

High variability arrivals

24

The G/G/1 Queue

If (1) < , (2) the distributions of job processing and inter-arrival times are independent and identically distributed (iid), and (3) jobs are processed on a first come, first served (FCFS) basis, then average waiting time in the queue can be approximated by the “VUT” formula:

2 2

E[W ]

[ ]2 1

q

A S

V U t

C CE S

25

Example

CA = CS = 1 E[S] = 1 Case 1: = 0.50 E[W] = 2, E[N] = 1 Case 2: = 0.66 E[W] = 3, E[N] = 1.98 Case 2: = 0.75 E[W] = 4, E[N] = 3 Case 1: = 0.80 E[W] = 5, E[N] = 4 Case 1: = 0.90 E[W] = 10, E[N] = 9 Case 1: = 0.95 E[W] = 20, E[N] = 19 Case 1: = 0.99 E[W] = 100, E[N] = 99

26

Example

CA = 1 E[S] = 1 = 0.8 Case 1: CS = 0 E[W] = 3, E[N] = 2.4 Case 2: CS = 0.5 E[W] = 4, E[N] = 3.2 Case 1: CS = 1 E[W] = 5, E[N] = 4 Case 1: CS = 1.5 E[W] = 6, E[N] = 4.8 Case 1: CS = 2 E[W] = 7, E[N] = 5.6

27

Facilities should not be operated near full capacity.

To reduce time in system and WIP, we should allow for excess capacity or reduce variability (or both).

28

A Single Stage System with Parallel facilities

Servers (production facilities)

Queue (logical or physical) of

jobs

Arrival (release) of

jobs

Departure (completion) of

jobs

29

The G/G/m Queue

If (1) < m, (2) the distributions of job processing and inter-arrival times are independent and identically distributed (iid), and (3) jobs are processed on a first come, first served (FCFS) basis, then average waiting time in the queue can be approximated by the “VUT” formula:

2( 1) 12 2 ( / )[ ]

2

mA S

q

C C mE W

m

30

Increasing Capacity

Capacity can be increased by either increasing the production rate (decreasing processing times) or increasing the number of production facilities

31

Increasing Capacity

Capacity can be increased by either increasing the production rate (decreasing processing times) or increasing the number of production facilities

In a system with multiple parallel production facilities, maximum throughput equals the sum of the production rates

32

Dedicated versus Pooled Capacity

Dedicated system: m production facilities, each with a single processor with production rate and arrival rate

Pooled system: A single production facility with m parallel processors, with production rate per processor, and arrival rate m

33

Dedicated versus Pooled Capacity

Dedicated system:

2 2

[W ]2

A S

q

C CE

2( 1) 12 2 ( / )[ ]

2 ( )

mA S

q

C C mE W

m

Pooled system:

34

Pooling reduces expected waiting time by more than a factor of m

Pooling makes better use of existing capacity by continuously balancing the load among different processors

35

The M/M/1 Queue

36

GX/GY/k/N

A Common Notation

G: distribution of inter-arrival

times

X: distribution of arrival batch

(group) size

G: distribution of service times

Y: distribution of service batch

size

k: number of servers

N: maximum number of customers

allowed

37

Common examples

M/M/1M/G/1 M/M/k M/M/1/NMX/M/1GI/M/1M/M/k/k

38

Notation in the Book versus Notation in the Lecture

Notes CT (cycle time): E(W)

CTq (cycle time in the queue): E(Wq)

WIP: E(N)

WIPq (WIP in the queue): E(Nq)

u: U (=/)

ra:

te: E(S); ts: E(X);

ca: cA

ce: cS

39

A single server queue The distribution of inter-arrival times

is exponential (Markovian arrivals) The distribution of processing times is

exponential (Markovian processing times)

Assumptions

40

Distribution of Inter-arrival Times

0 0

( ) : density function for the time interval between any two

successive arrivals

( ) , 0

1[ ] ( )

Pr( ) ( )

t

t

t T

T T

f t t

f t e t

E A tf t dt t e dt

t T f t dt e dt e

41

The Memoryless Property

( )Pr( )Pr( | ) Pr( )

Pr( )

T hh

T

t T h et T h t T e t h

t T e

42

The Taylor Series Expansion

2 3( ) ( )Pr(no arrivals in any interval of length ) 1 ...

2! 3!

Pr(no arrivals in any interval of length ) 1 (when is small)

Pr(a single arrival in any interval of length )

hh h

h e h

h h h

h h

(when is small)h

43

Exponential Inter-arrival Times and the Poisson Process

( )Pr( arrivals in an interval of length ) , =0,1,2...

!

n TT en T n

n

Poisson distribution

44

Distribution of Processing Times

0 0

( ) : density function for the time to process any job

( ) , 0

1[ ] ( )

t

t

g t t

g t e t

E S tg t dt t e dt

45

Similarly, when h is small,

Pr(processing time is not completed in interval of length ) 1

Pr(processing time completes in interval of length )

h h

h h

46

The Distribution of the Number of Jobs in the System

( ) Pr( ( ) ), probability that customers are in

the system at time nP T N T n n

T

47

The Distribution of the Number of Jobs in the System

1

1

( ) Pr( ( ) ), probability that customers are in

the system at time

P ( ) ( )(1 ) ( ) (1 )(1 ) ( )

( )( ) ( ) (1 )( ) ( )

n

n n n

n n

P T N t n n

t

T h h h P T h h P T

h h P T h h P T

48

0 1 1

00 1

P ( ) ( ) Plim ( ) ( ) ( ) ( ), for >0

P( ) ( ), for =0

n n nh n n n

T h P T dP T P T P T n

h dT

dP T P T n

dT

49

1 1

0 1

If 1, we can show that lim P ( ) , where is a constant

0

0 ( ) , for 1,2,3,...

0 , for =0

T n n n

n

n n n

T P P

dP

dT

P P P n

P P n

50

10 2

3

0 1

1 2 0

State rate out of state rate into state

0

1 ( )

2

j j

P P

P P P

2 3 1

1 1

( )

1 ( ) n n n

P P P

n P P P

The Birth-Death Model

51

1 0 0 0

00 00 0

( )

1 =1- 1

(1 )

nn

nnn n

nn

P P P P P

PP P P

P

52

0

0

20

20

2

: number of customers in the system (in the long run), a random variable

( ) Pr( )

(1 )

1

Var( ) Pr( )

(1 )

(1 )

n

nn

n

nn

N

E N n N n

n

N n N n

n

53

1

0

2

Pr( ) Pr( )

1 Pr( )

( ) ( )

1

1

n s

s

n

s

q

N s N n

N n

E N E N

54

( )( )

(1 )

1

E NE W

2

( )( )

(1 )

( )(1 ) 1

qq

E NE W

E S

Applying Little’s law

55

The M/G/1 Queue

A single server queue The distribution of inter-arrival times

is exponential (Markovian arrivals) The distribution of processing times is

general

56

The M/G/1 Queue (Continued…)

2

2 2

1[ ] [ ]

2 1

1[ ]

2 1

[ ] [ ] [ ]

[ ] [ ]

Sq

Sq

q

q

CE W E S

CE N

E W E W E S

E N E N

57

The G/G/1 Queue Revisited

2 2 2 2

2 2 2

[ ] [ ]2 1 2

[ ]2 1

[ ] [ ] [ ]

[ ] [ ]

A S A Sq

A Sq

q q

q

C C C CE W E S

C cE N

E W E W E S

E N E N

58

The M/M/m Queue

A queue with m servers The distribution of inter-arrival times

is exponential (Markovian arrivals) The distribution of processing times is

exponential (Markovian arrivals)

59

The Balance Equations

0 1

1 1

1 1

, for =0

( 1) =( ) for 1

=( ) for n n n

n n n

P P n

P n P n P n m

P m P m P n m

Using analysis similar to the one for the M/M/1 queue:

60

The Birth-Death Model

0 0 1 1

1 1 1 2 2 0 0

State rate out of state rate into state

0

1 ( )

2

j j

P P

P P P

2 2 2 3 3 1 1

1 1 1 1

( )

( )

n n n n n n n

P P P

n m P P P

10 2 3

61

The / / queue is a birth and death process with

if

if

n

n

M M m

n n m

m n m

62

0

0

01

0

, for 0!

, for !

1

[ ]! !( )

n

n

n

n n m

j mm

j

P P n mn

P P n mm m

P

j m mm

63

0Pr( )!(1 )

[ ] Pr( )

1[ ] Pr( )

, average utilization per server

m

q

q

N m Pm

m

E N N mm

E W N mm

Um

64

The G/G/m Queue

2( 1) 12 2

[ ]2

mA S

q

C C UE W

m

For a queue with a general distribution for arrivals and processing times, average time in the queue can be approximated as

65

Notation in the Book versus Notation in the Lecture

Notes CT (cycle time): E(W)

CTq (cycle time in the queue): E(Wq)

WIP: E(N)

WIPq (WIP in the queue): E(Nq)

u: U

ra:

te: E(S); ts: E(X);

ca: cA

ce: cS

66

Propagation of Variability

Single server queue:

Multi-server queue:

2 2 2 2 2(1 )

/D S AC U C U C

U

22 2 2 21 (1 )( 1) ( 1)

/( )

D A S

UC U C C

mU m

CD(i) = CA(i+1)i i+1

CS(i)CA(i)

67

Propagation of Variability

High Utilization Station

High Process Var

Low Flow Var High Flow Var

Low Utilization Station

High Process Var

Low Flow Var Low Flow Var

68

Propagation of Variability (Continued…)

High Utilization Station

Low Process Var

High Flow Var Low Flow Var

Low Utilization Station

Low Process Var

High Flow Var High Flow Var

69

Variability Relationships

, CD2 , CA

2

E(S), CS2

2 2

Queue Time

[ ] ( )2 1

A Sq

C C UE W E S

U

2 2 2 2 2

Flow Variability

(1 )

/D A SC U c U C

U

Processing

Time

( )E S

70

If utilization is low, reduce arrival arrival variability; if utilization is high, reduce process variability.

Operations with the highest variability should be done as late as possible in the production process.

71

A Production Line

N: number of stages in the production line Si: processing time in stage i (a random

variable), i=1,…, N U(i): utilization at stage i CA(i): coefficient of variation in inter-

arrival times to stage i

CS(i): coefficient of variation in

processing time at stage i

72

Time in System for a Production Line

E[Wq(i)]= V(i) U(i) E[S(i)]

2 2( ) ( )

2A SC i C i

( )[ ( )]

1 ( )

U iE S i

U i

1 1[ ] [ ( )] [ ( )] [ ( )]

N N

qi iE W E W i E W i E S i

( ) [ ]ii

U i E S

73

Time in System for a Production Line

2 2( ) ( )

2A SC i C i

( )[ ( )]

1 ( )

U iE S i

U i

( ) [ ]ii

U i E S

2 2 2 2 2( 1) (1 ( ) ) ( ) ( ) ( )A A SC i U i C i U i C i

74

Reducing Time in System (Cycle Time)

2 2( ) ( )

2A SC i C i

( )[ ( )]

1 ( )

U iE S i

U i

Reduce Variability • failures• setup times• uneven arrivals• process control• worker training

Reduce Utilization • arrival rate (yield, rework, etc.)• processing time (processing speed, availability)• capacity (number of machines)

75

Expected WIP in System for a Production Line

1

1

1

[ ] [ ]

[ ( )]

[ ( )]

[ ( )] ( )

N

i

N

i

N

qi

E N E W

E W i

E N i

E N i U i

Recommended