Applied Statistics (2)

Queuing Processes
Many phenomena for which mathematical description are desired involve
waiting lines either of people or material. A queue is a waiting line and
queuing processes are those stochastic process arising from waiting line
phenomena.
For example: The modeling of the arrival process of grain trucks to an
elevator, the utilization of data processing service at a customer center,
and the jobs at a job shop facility all involve waiting lines.
A queuing process involves the arrival of customers to a service facility
and the servicing of those customers. All customers that have arrived but
are not yet being served are said to be in the queue. The queuing system
includes all customers in the queue and all customers in service.



Queue
Server

System

Figure represents a queuing system with a mean arrival rate of  and

mean service rate of  , four customers in the system and three in the
queue.
Basic characteristics:
The basic characteristics of a queuing system are as follows :
(i) The input or arrival pattern of customers
(ii) The pattern of service
(iii) The number of servers or service charge.
(iv) The capacity of the system and
(v) The queue discipline.
 2

(i) The input pattern means the manner in which the arrivals occur. It
is specified by the interarrival time between any two consecutive
arrivals.
(ii) By the pattern of service, we mean the manner in which the service
is rendered. It is specified by the time taken to complete a service.
The time may be constant or it may be stochastic. If it is stochastic,
the pattern specification involves distribution of service time of a
unit.
(iii) A system may have a single server or a number of parallel servers.
An arrival who finds more than one free server may choose at
random any one of them for receiving service. If he finds all the
servers busy, he joins a queue common to all servers.
(iv) A system may have an infinite capacity- that is, the queue in front
of the serves (s) may grow to any length. Against this there may be
limitation of space is filled to capacity, an arrival of space is filled
to capacity, any arrival will not be able to join the system and will
be lost the system. The system is called a delay system or a loss
system, according to whether capacity is infinite or finite.
(v) The queue discipline indicates the manner in which the units are
takes for service. The usual queue discipline is first come, first
serve (FCFS) (first in first out FIFO) through sometimes there are
other service discipline, as last come first serve (which happen
sometimes in case of message) or service in random order.
Performance measure :
(i) Distribution of the number in the system at time t (the number
in the queue and the one being served, if any) is called the
queue length of the system at time t. By the number in the
queue mean the number of customer in the queue.
(ii) Distribution of the waiting time in the queue, the time that an
arrival has to wait in the queue.
(iii) Distribution of the virtual waiting means the length of time an
arrival has to wait had he arrived at time t.
 3

(iv) Distribution of the busy period being the length of time during
which the server remains busy period is the interval from the
moment of arrival of a unit at an any system to the moment that
the channel becomes free for the first time. The busy period is a
random variable.
Basic queuing theory rotation :
C: Number of identical servers.
L: Expected steady state number of customers in the queuing system,
E(N)
Lq: Expected steady state number of customers in the queue, E (Lq) Does
not include those receiving service, E(NS)
Ls: Expected steady state number customers receiving service, E[Ls]
 : Mean arrival rate of customer in the system
 : Mean service rate per server, that is the mean rate of service
completions while the server is busy.
N[t]: Random variable describing the number of customers in the system
at time t.
N: Random variable describing the steady state number of customers in
the system.
Nq[t]: Random variable describing the number of customers in the queue
at time t.
Nq: Random variable describing the steady state number of customers in
the queue.
Ns[t]: Random variable describing the number of customers receiving
service at time
Ns: Random variable describing the steady state number of customers in
the service facility.
pn[t] Probability there are n customer in the system at time t.
pn : Steady state probability there are customers in the system.
q: Random variable describing the time a customer spends in the queue
before service begins.
 E  Ns 
p: Server utilization = 
c c
 4

1
s: Random variable describing the service time E s  

1
 : Random variable describing interarrival time E   

w: Random variable describing the total time a customer spends in the
queueing system. w = q+s
w: expected (mean or average) steady state time a customer spends in the
system. W=E(w)= Wq+ Ws
Wq: Expected (mean or average) steady state time a customer spend in
the queue. Wq= E(q)= W-Ws
Ws: Expected (mean or average) customer service time. E[s]
# A basic notation called the Kandall notation, after David Kandall its
originator or has been developed to describe queueing system. The
notation has the form A/B/C/K/ m/Z, Where
A: describes intertribal time distribution
B: Describes the service distribution
C: number of servers
K: the system capacity (maximum number of customers allowed in the
system)
m: describe the number in population or source.
Z: describe the queue discipline.
Usually the shorter notation A/B/C is used and it is assumed that there is
no imit to the length of the queue, the customer source is infinite, and the
queue discipline is FCFS.
# Model M/G/1, has general service time distn Model M/M/1, queuing
system would give more information than M/G/1 Model.
# Example: Full Kandall notation M/E4/3/2/  / SIRO.
In this system the interarrival time is exponential, service time is Erlang-
4 for each of the three service, the maxim, system capacity 20 (3in
service and 17 in the queue), the source is infinite and the queue
discipline is service in random order.
# For M/G/1- model only some of the moment need to be known to
complete useful performance information.
 5

WS
# Traffic intensity, a = , it is also known as offered load.
E r 
Ws: is the average service time per service.
E(r) : is the average interarrival time for all customers.
1 1
# As   and   , the traffic intensity can also be written as
E r  WS

WS or  /  . The quantity  = a/c= is called the server utilization,
MC
because it represents the average fraction of the time that each server is
busy.
# W: average customer time in the system.
Wq: average customer queuing time.
th
 w [90] : 90 percentile value of w.
th
 q [90] : 90 percentile value of q.
L: average number of customers is system.
Lq: average number of customers in queue.
pn: probability there are n customers in the queueing system.
# Little’s Law:
One of the foundation of queueing theory is the formula.
L  W (1)
The formula (1) applies to any system in equilibrium in which customers
arrive, spend a certain amount of time, and then depart. In Ea. (1)  used
for average arrival rate, W used for the average time a customer spends in
the system, and L used for the average number of customers in the
system.
Rate- Equality Principle:
The rate at which a process enters a state n  0 equals the rate at which
the leaves that state n. In other words the rate entering and the rate of
leaving a particulan state are same for every state, that is.
rate in= rate out or rate up= rate down, for every state.
The simple M/W/1 queue :
 6

 The two Ms refer to the fact that both the inferarrival and the
service distribution are exponential (and that memoryless or
Markovian and 1 to the fact that there is a single servers).
 In such a queueing system, the arrivals occur from an infinite
source in accordance with a poison process with parameter  , that
1
is, the interarrival times are independent exponential with mean ,

the service times are independently and exponentially distributed
with parameter say  and there is only one server. The queue

discipline is FCFS; the utilization factor is P  =a,  and  being

the arrival and service rates respectively.
Steady- state solution of M/M/1
Let x(t) denote number of customers in the system at time t and define pn,
n  0 by
pn  lim P  X t   n  1
t 

For each n  0 , the rate at which the process enters state n equals the rate
at which it leaves state n. Consider first state 0. When in state 0 the
process can leave only by an arrival the system is empty. Since the arrival
rate is  and the proportion of time that the process in state 0 is p0. It
follows that the rate at which the process leaves the state 1 is  p0 . On the
other hand state 0 can only be reached from state 1 via diparture. That is
if there is a single customer in the system and he completes service, then
the system becomes empty. Since the service rate is  and the proportion
time that the system has exactly one customer is p1. It follows that the
rate at which the process enter state 0 is  p1. Hence from rate equality
principle, we get
p0  p1
Now consider state 1 : The process can leave the state either an arrival
(which occur at rate or departure rate  ) or departure (which occurs at
rate  ) Hence, when in state 1, the process will leave this state at rate
 7

   . Since the proportion of time the process is in state 1 is p 1, the rate

at which the process leaves state 1 is     p1
On the other hand, state 1 can be entered either from state 0 via an arrival
or from state 2 via a departure. Hence the rate at which the process enters
state 1 is p0  p2
Therefore
For state 0 : p0  p1
For state n , n 1 :     pn  pn 1  pn 1
Eq (2) is renown as balance equation.
From first equation of (2) we get

p1  p
 0
and from second equation of (2) we get
   
pn 1  pr   pn  pn 1  ; n  1
   
Solving interms of p0 yields ,
p0  p0

p1  p0

   
p2  p1   p1  p0 
   
2
     
 p2   p0  p0  p0     p0
     
   
p3  p2   p2  p1 
   
 
2
  2 0   2 
   p0    p    p0 
      
3

 p3    p0

   
p4  p3   p3  p2 
   

3
   3
   
3

   p0    p0    p0 
       
4

 p4    p0

 8

Proceeding in the similar manner on can find,

n 1

pn 1   p0

To determine p0 Let
n



1   pn     p0
n0 n0   

     2   3 
 1         ..... p0
        
1 
 p ;| | 1
 0 
1


 p0  1 


Hence, p0  1 

    
pn  1     n; n  1
    
 
Note : > 1, the all Pn would be 0. Hence we assume < 1 i.e   
 
So there is no limiting probability if . Since customers arrive at a poisson
rate  , expected total arrival number of arrivals by time t is  t.
If there where always customers present then the customers served could
be a poisson process having rate  , since the time between successive
1
services would be independent exponentials having mean . Hence the

expected number of customers served by time is no greater that  t; and
therefore the expected number in the system at time t is at least
t  t      t . Now if    , then the preceding number goes to

infinity as becomes large. That is, 1, the queue size increase without

limit and there will be no limiting probabilities. Note that the condition

 1 is equivalent to the condition that the mean service time be less

than mean time between successive arrivals. This is the general condition
 9

that must be satisfied for limited probabities to exist in most single server
queuing system.
Some performance Measures.
(i) The average number of customers in the system in steady state is given
by
L  EL

  npn
n0


  
  n   1  
n0     


 n 1   
n0
n
;


 1     np n
n0

  1     np n 1
n 1


  1    1  2   3 2  4  3  ................. 
  1  P 1   
2
1  x  2

 1  2 x  3 x  .......


1 
/

1  / 

 L
 
(i) Note: If N(t) denotes the number of customer arrivals by time t, then
N t 
a  lim
t t

 a indicates the average arrival rate of entering customers.

(ii) The average amount of time a customer spends in the system is given
by
L
W 

1 
 
  
1
W 
 
 10

(iii) The average amount of time a customer spend waiting in queue is

given by

L
W

1 

  
1
W 
 
# E(S) is the average amount of time a customer spends in service.
(iv) The average number of customers waiting in the queue is given by
Lq   Wq
 2
 =
         

# Variance of the average number of customers in the system.

  
We have from (i) E L    , 
   1  

 

E L2   n 2  n
n0


  n 2 1    n ; 
n 0 

 1    

n 2
 
 n  n n
n 0

 2 2  
 1      2
 1    1    
3

2 2 
 
1    1  
2

  2

1   2
Var L   E L2  E L    2

  2 2
 
1   2 1   2
P

1   2
* What is the necessary condition for existences of steady states?
 11

* When  1, EL and Var L tend to  also 0  0, n  0, n  1, so

that probability that the system is empty or contains a certain fixed
number is very, very small.
* The large variance of the number in the system implies that a randomly
observed system size is likely to be very different from the expected
system size.
* Show that queue length distributions under both the FCFS and service
in Random order (SIRO) discipline will be same.
* The state transition rate diagram M/M/1 system

  
State : n-1
0 1 2 n

   

P (Server busy) = 1-  (L=0)

= 1- (1-  )
= 

=

Explain : W= Wq+Ws or Meaning the above relation means that, on the
average, customers who must wait will wait in the queue one average
service time longer than the average customers waits.
P-1 Traffic to a message swit-ching career for Extraterrestial
communications Corporation arrives in a random pattern(Means
exponential interarrival) at an average rate, of 240 message per minimum.
The line has a transmission rate of 800 characters per second. The
message length distribution is approximately exponential with average
length of 176 characters. Calculate the principal statistical measures of
the system performance assuming that a very large number of message
 12

buffers is provided. What is the probability that 10 or more message are

waiting to be transmitted?
Ans : Average message length = 176 char
line speed = 800 char.
message = 240 per min
The average service time i.e the average time required to transmit a
message.
average message length
Ws =
line speed
176
= sec
800
=0.22
240 240
Arrival rate,     4 mess / sec
1min 60
Server utilization, P = Ws   = 4  0.22 = 0.88
Therefore the communication line is transmitting outgoing message 88%
of the time.
Performance Measures:
Average number of message in the system.

L  7.33 Mess
1 
Average number of messages in queue,
2
Lq   6.45 mess
1 
Average response time,
Ws
W  1.83 Seconds
1 
Average queuing time,
Wq   W 1.61 Seconds
th
90 Percentile time in the system,
 w 90 W ln 10   4.21 Seconds
90th Percentile time in the system,
 q 90 W ln 10    3.98 Seconds
 13

Since 10 or more messages are queuing or more messages are in the

system. So, the probability that 10 or more message are waiting =  11
= (0.88)11
= 0.024508
* What would be the average response time if the traffic rate into the
center increases by 10%
10
240  240 
Ans:   100  4.4 mess
1 min
 P  Ws  4.4  0.22  0.968
Hence average response time
Ws 0.22
W    6.875
A   2  0.9868

P-2 : A branch office of a large engineering firm has one on-line terminal
connected to central computer system for 16 hours each day. Engineers,
who work through out the city drive to the branch office to use the
terminal for making routine calculations. The arrival pattern of engineers
is random (Poisson) with an average of 20 per day using terminal. The
distribution of time spend by a engineer at the terminal is exponential
with an average time of 30 minutes. Thus the terminal is 5/8 utlized
 1 
 20  10 hour out of 16 hour availeable  . The branch manager receives
 2 
complaints form the staff about the length of time many of them have to
wait to use the terminal. It does not seem reasonable to the manager to
procure another terminal when the present on is only used five-eight of
the time, on the average. How can queuing theory help this manager?
Ans: The M/M/1 queueing system is reasonable model with  = 5/8 as we
comput above Ws= 30 minutes.
Average time engineer spends at the branch office,
Ws
W 
1 
30

1 5 / 8
 14

= 80 Minutes
Average number of engineers waiting the queue,

Lq 
2

5 / 82 1.0417
1  1 5 / 8
Average number of engineers in moment queue =
1 1
 8 / 3
1 P 1 5/8
Average waiting time in queue,
PWs 78  30
Wq  
1  1 5 / 8
5 8
   50 Minutes
8 3
Average waiting time of those who must wait
=W
= 80 Minutes
90the percentile of time in the queue,
 q 90   W lnP  146.61 Minutes
90th Percentile of time in the branch office
 w 90   23W  2.3 80 184 Minutes
Since  = 5/8, one three-eights of the engineers who use the terminal
need not wait. For those who must wait, the average wait for the terminal
80 minutes, quite a long wait, Ten percent of the engineers spend over. 3
hours, actually 184 minutes in the office to do an average of 30 minutes
of computing. The probability of waiting more than an how to use the
terminal is
60
5 
P q  60   e 80  0.295229
8
or atmost 30%
These results may seem little starting to those not acquainted with
queuing theory. It might seem, institives that, adding another terminal
would cut the average time in half, from 50 minutes to 25 minutes (to 40
minutes for those who must wait). The queueing theory we have
presented so for should suffice to convince the manager that an
improvement is needed.
 15

1
P-3: If the arrival and departure rates and M/M/1 queue are per minute
2
2
and per minute, respectively find the average waiting of a customer in
3
the queue.
Ans:
1
 | min
Given, 2
  2 3 | min
The average time of waiting of a customer in the queue, Wq=
 1
 2
= 4.5 minute.
     2
   2
1
2
3
P-4: Arrivals at a telephone both are considered to be poisson with an
average time of 12 minute between one arrival and the next. The length of
a phone call is assured to be distributed exponentially with mean 4
minutes.
(i) Find the average number of persons waiting in the system.
(ii) What is the probability that a person arriving at the both will have to
wait in queue.
(iii) What is the probability that it will take him more that 10 min,
altogether to wait for the phone and complete his call?
iv) Estimate the fraction of the day when the phone will be inuse.
v) The telephone department will install a second both, when convinced
that an arrival has to wait on the average for at least 3 min. for phone. By
how much the flow of arrivals should increase in order to justify should
increase in order to justify a second both?
vi) What is the average length of the queue that?
1
Ans: Mean of exponential distribution =

1 1
 12, i.e   Min
 12
Given : Service follows exponential distribution
1
 Mean   4 min

 16

1
 Mean service rate,   min
4
Performance Measure :
(i) The average number of persons waiting in the system,
 1
Ls   12
 0.5 Customer
  1
4  112
(ii) The Probability that a person at the both will have to wait in the
queue.
(This means that the queue is non-empty)
PLq  0  1  PLq  0
 1 0
 
1  1  
 
 1 1
  12

 1
4 3
(iii) The probability that the waiting time in the system exceeds 10 Min;
t = 10 min
 P Ws  10   e      t
1 1 
     10
e  4 12 

1
 10
e 6

5 3
e
 0.1889
iv) The fraction of the day the phone will be in use.
The phone will be idle when there is no arrives P (The phone will be idle)
= po

 1

1
1  12
1
4

 2
3

 1
P(The phone will be in use) = 
 3
The fraction of the day when the phone will be in use= 1
3
 17

V) The second phone will be installed if Wq  3 . That is the arrivals have

wait on the average for at least 3 min.
i.e. if Wq  3

and if 3
    

i.e if 3
11 
 
44 
3 1 
or,     
4 4 
3 3
or,    
16 4
7 3
or 
4 16
3
or  
28
Hence, to justify a second phone, the arrival rate should be increase by
3 1 1
  / Min
28 12 42

(vi) The average length of the queue that forms from time to time,
E (Lq / the queue is always availeable)


 
1
= 4

4  12
1 1

1
= 1
4

6
=
4
= 3/2

P-5. In a railway marshalling yard, goods trains arrive at a rate of 30

trains per day. Assuming that, the interarraival time follows an
 18

exponential distribution and the service time distribution is also

exponential with an average of 36 min calculate:
(i) The mean queue size.
(ii) The probability that the queue size exceeds 10.
(iii) If the input of trains increases to an average 33 per day, what will be
change in (i) and (ii)?
Ans. The arrival and service time follow exponential distribution.
Therefore it is an M|M|1 infinite capacity model.
Here
  30 trains/ day
1
 / min
36
1
=  60  24
36
= 40 trains/ day
Traffic intensity,
 30 3
=    4
 40
Performance measure :
i) The mean queue size (length),

Lq 
2

30 
2

3 3 9

     40 40  30 4 1 4
= 2.25
(ii) The probability that the queue size exceeds 10,
10 10
  3
PLq  10         0.056
   4
(iii) If the input (arrival) of trains increases to an average 33 per day, then
  33 trains/ day
33 33 33 33
 Lq      3.89 trains / day
40 40  33 40 7
The change in queue size = 3.89- 2.25
= 1.64 trains
10
 
Now PLq  10   
 
 19

10
 33 
 
 40 
= 0.146
So, the change in probability
= 0.146- 0.056
= 0.090

M|M|C:  / FIFO)
Multi-server posson queueing model.
The queueing system deals with queues which are being served by
parallel service chan in which each server has an independent and
identically distributed exponential service distribution. The arrival is
consider be poisson. This model has the number service channel is c.
In the case of c servers working together we have two cases:
(i) If the number of customers in the & is less than c, i.e n<c, then
only n of the c servers will be busy and hence the mean sens
rate will be n  .
(ii) The number of customers in the is more than c, i.e. n>c then all
the c servers will be busy and hence the mean service rate will
be c  , i.e
 n ; o  n  c 
Mn   
c ; n  c 
and n  
n
1 
p n    p o ;1  n  c
n!   
n
1 
Pn  n 1   Po ; n  c
C 

system busy = server busy =  = is also called traffic intensity
c
(M|M|C- Model)
Performance Measures (M|M|C- Model)
1. The average number (expected number) of customers in the queue
(Lq):
 20

When all e-server busy-

C c  c 1 
Lq= o ;  
C! (1   ) 2
C
2. The average numbers of customers in the system (Ls):

 C c
P c 1
Ls= Lq + = Po  
 C! 1  P  2

3. Average time (expected waiting time) : a customer spends in the
system (ws):

Ls C c  c 1 1
Ws   o 
 c! 1   2

4. Average time (expected waiting time) : a customer spends in the queue
(Wq):
Lq C c c  1
Wq   o
 C! 1   2
5. The probability that an arrival has to wait as there are servers an arrival
has to wait. There are more than C customers in the day.
Cc C
P Ws  0   P n  C   P0
C1 1 
6. The probability than an arrival has to the service without waiting,
P(arrival getting the service without waiting = 1-P (arrival has to
wait)
C c C
=1 0
C! 1  
7. The probability that some one will containing.
C c  c 1
Pn  c  1  po
C! 1  
Cc c 0
Pn  c   p
C! 1  
8. The mean waiting time in non-empty queue E
1
E Wq /Ws  
 1   
 21

9. The average number of customers (in non- simple queue) who have to
actually wait.
P
Lq 
1 P
P-1 What is the probability that arrival to an infinite capacity server
poisson queueing system with  /  = 2 and po = 1
9 enter the
service without waiting?
Ans: There are 3 cervess.
 It is an M|M|C|  Model.
Given :
 1
 2 and po 
 9
An arrival enters the service without waiting while the number of
customers getting in the system are less than 3.
P(Without waiting) = p(n< 3 ) = po+ p1 + p2
n
1   o
pn =   P ; n  c, c  3
n!   
1 2 1 1
 p n  3     4 
9 9 2 9
5
=
9
P-2: If there are 2 servers in an infinite capacity poisson queue system
with  per hour, what is the percantage of the idle to for each
server?
Ans: It is an M|M|C|  model with c= 2,   10 / hr,   15 / hour
 10 1
P   
c 30 3
C  2
3

 
 C 1 C   C  
n c 1
 C 1  3    3   = 1
2 n 2 c
Po     
C!1      no  1
2!1   
 no n! n! 2
  3 
 The percentage of idle time for each server 50%