Queues appear in many aspects of life.
Some are clearly visible, as in the queues at
supermarket check-out tills: others may be less
obvious as in
information telephone lines, or
in hospital waiting lists.
In the latter an individual waiting may have no idea
how many persons are in front of him or her in the
hospital appointments list.
Generally, we are interested in the long-term behavior
of queues for future planning purposes
—does a queue increase with time or
does it have a steady state, and if it does have a steady
state, then, on average, how many individuals are there
in the queue and
what is the mean waiting time?
Imagine an ATM machine
there are some customers here in line in a queue
every once in a while, a new customer arrives and
when ATM is empty then the first customer in the line will be served
He spend some time at the ATM and then he leaves
then the next customer is served
of course, new customers keep coming
there can be multiple ATM here so as long as one ATM
is empty customer can be served at the same time in
parallel
We want to modeling this queue
The number of customers in a queue are modeled by
the Poisson Process
The probability that there are n customers in the queue
(include the customer being served at the server). this is
modeled by the Poisson distribution
The time between two arrivals two consecutive arrivals
of customers
is modeled by the exponential distribution
The time spent at the server is modeled by exponential
distribution
The density function of the time spent at a server
The problem of a single server queue
There's one ATM
the probability that a new customer arrives
this probability the customer will leave from the server during
we model the number of customers in the q
including the one being served
by the random variable n
N(t)
Based on these assumptions we are going to set up a
set of differential difference equations
the probability that there are n customers at time t,
in the queue
there are no customers at time
We want to find, there are no customers at time
First state is: there are no customers at time
So, suppose there are no customers at time t , the
probability is
And, should be no new customers arriving
So, the probability will be
Suppose, there are no customer being served because that number of customers zero
so no customer will leave during that time
Or
if there are only one customer then that customer must be leave during de
so that probability is
So we have
in general, suppose there are n customers at time t
plus delta t
the possible scenario at time t, n customer, n-1
costumer or n+1 customer
N customer at n customer at t
no customer leaves
no new customer arrives
n customer at t
one customer leaves
one customer arrives
( + )
N-1 customer at t
the new customer should arrive
and no customer should leave 1-
N+1, customer at t
no new customer arrives
one customer leave
So
let's rearrange this different
difference equations
Limiting Processes
In the limiting process we look at the long-term
behavior of the queue
Assuming that it has a limiting state. Not all queues
do:
if rate of arrivals is large compared with the service
time then we might expect the queue to continue to
grow without bound as time progresses.
For the moment let us assume that there is a limiting
state, that
λ < μ, and that we can write
The probabilities now approach a constant distribution which does not depend
on time. In this case it is reasonable to assume that
Equations now become
This is a second-order difference equation. It is easy to solve iteratively
since
and, with n = 1, in
We have
With substituting
We have
The next iterate is
and, in general,
and at this point the value of p0 is unknown
so how can we determine this value of p 0
before determining the value of p 0 let us define a
quantity called the traffic density
traffic density
ρ=
Actually if we define
as the service time and
I as the inter arrival time of new customers
then this traffic density is this expectation value of the
over expectation value of inter arrival time
pn is expressed as
this is probability mass function of the number of customers
the sum of all pn and ranging from 0 to infinity should
add up to 1
but the geometric series on the left will only converge
if
Hence,
which is the probability mass function of a geometric
distribution with parameter
then the series does not converge and {pn} cannot represent a probability
distribution. In this case the queue simply increases in length to infinity as t → ∞
whatever its initial length. On the other hand if
and the rate of arrivals is less than the average service time with the
result that a steady state should be achieved as we might expect.
Notice that, if
the queue grows in length although there appears to be equilibrium’ between
arrivals and service.
(i) Server free. The probability that the server is free
when a customer arrives is
(ii) Length of queue. The mean length of the queue
(including the person being served) is, if N represents
a random variable of the number in the queue,
(iii) Waiting time. How long will a customer expect to
wait for service on arrival at the back of the queue, and
how long will the customer expect to queue and be
served?
A customer arrives and finds that there are n
individuals ahead including the person being served
the random variable representing
the time for the service of
customer i.
each of them will be served with
n for ti seconds
the sum of the random variables of service times of the first n queuing
customers.
These random variables are independent, each with an exponential
density function with the same parameter
the sum of independent identical distributed
exponential random variables follow a gamma
distribution
density function is given by
Notice that this is the density function for Sn which is
the waiting time given that there are n customers in
the queue
But usually we don't care about how many customers
are there , we only care how long we have to wait
We want to actually calculate the probability density
function for the waiting time irrespective of the
number of customers so let's say this a random
variable of waiting time irrespective of customers is S
So we want to first calculate the distribution function
for this this random variable
Using the law of total probability, the probability that S is greater than t is
given by
WE KNOW
we put all these together in this equation
Hence,
this distribution function of waiting time
irrespective of the number of customers
this equation means this is application of the partition
theorem so if there are n customers that the
distribution function is this
the density function of S is the derivative of
cumulative distribution function with respect to time t
Cumulative distribution function is
< t)
The associated density function is
Finally, the expected value of S is
With n customers ahead the next customer could expect to wait for a time
to reach the server. Until service is completed the customer could expect
to spend time
the horizontal axis is the time from some point and
the vertical axis is the number of customers so this
shows the evolution of the number of customers after
some point in time
at first there are no customers so which is called the
slack period
at time t one customer arrives so now the system
enters the busy period so if the server is occupied
with at least one customer it is in the busy period
now we are interested in the average duration of the
busy period
we represent each slack period by the duration of each
slack period by s i s1 s2 s3 and so on and the duration
of the busy periods are represented by b1 b2 b3 and so
on
and periods when the server is busy, namely
The times t1, t3, t5, . . . are the times when a new
customer arrives when the server is free, and
t2, t4, t6, . . . are the times at which the server becomes
free.
The periods denoted by b1, b2, . . . are known as busy
periods.
A question whose answer is of interest is: what is the
expected length of a busy period?
what is the expected length of a busy period?
Suppose that the server is free at time t = t2.
The arrivals form a Poisson process with parameter λ,
The expected time until the next customer arrives is
1/λ, (the density function is exponential. )
The average lengths of a large number n of slack
periods should approach 1/ λ.
: the probability that the server is free
the probability that the server is busy
:
Hence the ratio of the average lengths of the slack
periods to that of the busy periods is
s/b =
s = b
which is the mean length of the busy
periods
The mean length of the busy periods
b
Server free. The probability that the server is free when
a customer arrives is
p0 = 1 − ρ.
Length of queue. The mean length of the queue
(including the person being served) is, if N represents
a random variable of the number in the queue,
traffic density
S and I are independent random variables of the
service and inter-arrival times, respectively. When S
and I are both exponentially distributed, then
E(S) = 1 /µ and E(I) = 1/λ
ρ = E(S)/ E(I) ,
Waiting time
Queues with multiple servers
In many practical applications of queueing models,
there is more than one server
in banks, supermarkets, and hospital admissions.
1 2 3 r
we consider a single queue so there is only one queue
and as a server becomes free then customer is served at
the server which server doesn't matter but as long as
it's free then a customer is served
we also consider the service time follows the
exponential distribution
the number of customers including those that are
being served is represented by this random variable n
of
Suppose that the queue faces r servers.
the arrivals form a Poisson process with parameter λ
the service time at each counter has exponential
distribution with rate μ
pn(t) be the probability that there are n people in the
queue at time t including those being served
If all counters will be occupied by a
customer
If some counters will be free
Each server has the probability of becoming free with
There are r servers that are being so the total
probability should be
the probability of busy server
we can derive the following incremental equations:
this is the same as the single server case
this is the case where
what is different from
the single server
With the usual limiting process. we obtain the time-
dependent equations for the queuing process
Assuming that the derivatives tend to zero as t → ∞(as
in the previous section), the corresponding limiting
process (if it exists) for the queue with r servers is
There are three main factors which characterize
queues:
the probability distributions controlling arrivals,
service times, and
the number of servers.
arrival distribution, G1/ service distribution, G2 /
number of servers, n
If G1 and G2 are exponentials with λ and µ
(that is, both are Poisson processes),
then both processes are Markov with parameters λ and
µ, respectively
The Markov property means that the probability of the
next arrival or the probability of service being
completed are independent of any previous
occurrences.
We denote the processes by M(λ) and M(µ)
. the single-server queue is denoted by
M(λ)/M(µ)/1, and
the n-server queue by
M(λ)/M(µ)/n.
single-server queue with Markov inter-arrival but fixed
service time τ
M(λ)/D(τ)/1, where D stands for deterministic.
If the service time for a single-server queue has a
uniform distribution
M(λ)/U(µ)/1
service time for a single-server queue has a uniform
distribution with density function
µ 0 ≤ t ≤ 1/µ,
0 elsewhere,
