Traffic Engineering

C. U. Shah College of Engineering & Technology, Wadhwan city

Introduction
Switching elements are not the only parameters which decides
the blocking probability of a network.
There are many parameters like digit receivers, interstage
switching links, call processors and trunk between exchanges
along with, which decide the blocking probability of a
network.
Traffic engineering analysis enables one to determine the
ability of telecommunication network to carry a given traffic at
a particular loss probability.
It provides a means to determine the quantum of common
equipment required to provide a particular level of services for
a given traffic pattern and volume.
2
Network Traffic Load & Parameters
3
busy hour (BH)
Network Traffic Load & Parameters
In a day, the 60 minute interval in which the traffic is the
highest is called the busy hour (BH).
1) Busy Hour : Continuous 1 hour period lying wholly in the
time interval concerned, for which the traffic volume or the
number of call attempts is greatest.
2) Peak Busy Hour : The busy hour each day; its usually
varies from day to day, or over a number of days.
3) Time Consistent Busy Hour :The 1-hour period starting at
the same time each day for which the average traffic volume or
the number of call attempts is greatest over the days under
consideration.
For ease of records, the busy hour is taken to commence on the
hour or half-hour only.
4
Network Traffic Load & Parameters
A call attempt is said to be successful or completed if the
called party answers.
Call Completion Rate (CCR) is the ratio of the number of
successful calls to the number of call attempts.
The number of call attempts in the busy hour is called busy
hour call attempts (BHCA), which is an important parameter
in deciding the processing capacity of a common control or a
stored program control system of an exchange.
Networks are usually designed to provide an overall CCR of
over 0.70.
A CCR value of 0.75 is considered excellent and attempts to
further improve the value is generally not cost effective.
5
Network Traffic Load & Parameters
A related parameter that is often used in traffic engineering
calculation is the busy hour calling rate, which is defined as
the average number of calls originated by a subscriber during
the busy hour.
Another useful information is to know how much of the days
total traffic is carried during the busy hour is day-to-busy
hour traffic ratio which is the ratio of busy hour calling rate
to the average calling rate for the day.
Typically, this ratio may be over 20 for a city business area
and around six or seven for a rural area.
6
Example 1
An exchange serves 2000 subscribers. If the average BHCA is
10,000 and the CCR is 60%, calculate the busy hour calling
rate.
Answer :
Average busy hour calls = BHCA x CCR
= 6000 calls.
Busy hour calling rate =
Average busy hour calls/ Total number of subscribers = 3.
7
Network Traffic Load & Parameters
8
For analytical treatment, all the common subsystems of a
telecommunication network are collectively termed as servers.
In other publications, the term link or trunk is used.
The traffic on the network may then be measured in terms of
the occupancy of the servers in the network.
Such a measure is called the traffic intensity which is defined
as
Where A
0
is dimensionless and called Earlang (E).
0
Period for whicha server is occupied
A
Total period of observation
=
Example 2
9
In a group of 10 servers, each server is occupied for 30
minutes in an observation interval of two hours. Calculate the
traffic carried by the group.
Answer
= 30/120 = 0.25 E
So total traffic carried by the group = 10 X 0.25 = 2.5 E
0
Period for whicha server is occupied
A
Total period of observation
=
Example 3
10
A group of 20 servers carry a traffic of 10 Earlangs. If the
average duration of a call is three minutes. Calculate the
number of calls put through by a single server and the group as
a whole in a one hour traffic period
Answer
Traffic carried per servers =
10/20 = 0.5 E (i.e. server is busy for 0.5 x 60=30 minutes in
one hour)
Number of calls put through by one server =
Duration for server is busy in one hour/ Average call duration
= 30/3 = 10 Calls.
So total number of calls put through by the group =
= 10 X 20 =200
Network Traffic Load & Parameters
Traffic intensity is also measured in another way. This
measure is known as centum call second (CCS) which
represents a call-time product.
One CCS may mean one call for 100 seconds duration or 100
calls for one second duration each or any other combination.
CCS as a measure of traffic intensity is valid only in
telephone circuits. For the present day networks which support
voice, data and many other services, erlang is a better measure
to use for representing the traffic intensity.
Sometimes, call seconds (CS) and call minutes (CM) are
also used as a measure of traffic intensity.
1 E = 36 CCS = 3600 CS = 60 CM.
11
Example 4
A subscriber makes three phone calls of three minutes, four
minutes and two minutes duration in a one-hour period.
Calculate the subscriber traffic in erlangs, CCS and CM.
Answer :
= (3 + 4 + 2)/ 60 = 0.15 E
Traffic in CCS = (3 + 4 + 2) x 60 / 100 =540/100 = 5.4 CCS
Traffic in CM = 3 + 4 + 2 = 9 CM.
12
=


Network Traffic Load & Parameters
Whenever we use the terminology "subscriber traffic" or
"trunk traffic", we mean the traffic intensity contributed by a
subscriber or the traffic intensity on a trunk.
As mentioned above, traffic intensity is a call-time product.
Hence two important parameters that are required to estimate
the traffic intensity or the network load are
Average call arrival rate, C
Average holding time per call, t
h
And load offered to the network in terms of these parameter
A = Ct
h
C and t
h
must be expressed in like time units.
For example, if C is in number of calls per minute, t
h
must be
in minutes per call.
13
Example 5
Over a 20-minute observation interval, 40 subscribers initiate
calls. Total duration of the calls is 4800 seconds. Calculate the
load offered to the network by the subscribers and the average
subscriber traffic.
Answer :
Mean arrival rate C =
=40/20 = 2 calls/minute.
Mean arrival time t
h
=
=4800/ (40 X 60) = 2 min/call.
Therefore, offered load =
= 2 X 2 = 4E.
Average subscriber traffic =
=4 /40 = 0.1 E
14
Network Traffic Load & Parameters
We have calculated the traffic in two ways: one based on the
traffic generated by the subscribers and the other based on the
observation of busy servers in the network.
It is possible that the load generated by the subscribers
sometimes exceeds the network capacity.
If overload traffic is rejected than it is Loss system. i.e.
Automatic telephone exchange.
If overload traffic is delayed until the resources are available
then it is Delay system. i.e. operator- oriented manual
exchange.
In the limit, delay systems behave as loss system.
15
Network Traffic Load & Parameters
The basic performance parameters for a loss system are the
grade of service and the blocking probability, and for a delay
system, the service delays.
Average delays, or probability of delay exceeding a certain
limit or variance of delay may be important under different
circumstances.
The traffic models used for studying loss systems are known
as blocking or congestion models and ones used for studying
delay systems are called queuing models.
16
Grade of Service and Blocking Probability
In a loss system, the amount of traffic rejected by the network
is an index of the quality of the service offered by the network.
This is termed Grade of Service (GOS) and is defined as the
ratio of lost traffic to offered traffic.
Offered traffic is the product of the average number of calls
generated by the users and the average holding time per call.
A =Ct
h
Actual traffic carried by the network is called the carried
traffic and is the average occupancy of the servers in the
network
17
0
Period for whichserver is occupied
A
total period of observation
=
Grade of Service and Blocking Probability
And
Where A =Offered traffic
A
0
= carried traffic
A- A
0
= lost traffic
Smaller the value of GOS, better the service is.
In India GOS = 0.002, which means 2 calls in every 1000 calls
may be lost.
The GOS of the full network is determined by the highest
GOS value of the subsystems in a simplistic sense.
18
0
A A
GOS
A

=
Grade of Service and Blocking Probability
Blocking probability P
B
is defined as the probability that all
the servers in the system are busy.
In a system with equal number of servers and subscribers , the
GOS is zero as there is always a server available to a
subscriber.
On the other hand, there is a definite probability that all the
servers are busy as a given instant and P
B
is non zero.
The fundamental difference is that the GOS is a measure from
the subscriber point of view whereas P
B
is a measure from
network or switching point of view.
GOS is defined by observing the number of rejected subscriber
calls, where P
B
is defined by observing the busy servers in
switching system.
19
Grade of Service and Blocking Probability
GOS is called call congestion or loss probability and the P
B
is called time congestion.
In the case of delay system, overload traffic is queued, so GOS
is not a useful measure for the same.
So, delay probability, the probability that call experience
delay is a useful measure.
If queue lengthy becomes very large and system becomes
unstable, and easy way to stable the system is to operate
system as loss system until the queue is cleared.
This technique of maintaining the stable operation is called
flow control.
20
Grade of Service and Blocking Probability
Quality of service (QOS) is used in more recent times and it
includes the quality of speech, error-free transmission capacity
etc.
Summary
Subscriber viewpoint
GOS = call congestion = loss probability
Network viewpoint
Blocking Probability = time congestion
21
Modelling Switching System
Subscribers generate calls in a random manner, so call
generation is a random process.
A Random Process or a Stochastic Process is the one in
which one or more quantities vary with time in a such away
that the instantaneous values of the quantities are not
determinable precisely but are predictable with certain
probability. This quantities are called Random Variables.
Figure 8.2 shows typical fluctuations in the number of
simultaneous calls in a half-hour period. The Pattern signifies
a typical random process.
22
Modelling Switching System
23
Modelling Switching System
The values taken on by the random variables of a random
process may be discrete or continuous.
In the case of telephone traffic, the random variable
representing the number of simultaneous calls can take on only
discrete values whereas a random variable representing
temperature variations in an experiment can take on
continuous values. Similarly, the time index of the random
variables can be discrete or continuous. Accordingly, we have
four different types of stochastic processes:
1) Continuous time continuous state
2) Continuous time discrete state
3) Discrete time continuous state
4) Discrete time discrete state.
24
Modelling Switching System
A discrete state stochastic process is often called a chain.
Statistical properties of a random process may be obtained in
two ways:
Time Statistical Parameters, by observing its behaviour over
a very long period of time. Fig 8.3(a)
Ensemble statistical parameters, by observing
simultaneously, a very large number of statistically identical
random sources at any given instant of time. Fig 8.3(b)
25
Modelling Switching System
26
Modelling Switching System
Random processes whose statistical parameters do not change
with time arc known as stationary processes.
The random processes which have identical time and ensemble
averages arc known as ergodic processes.
In some random processes, the mean and the variance alone
are stationary and other higher order moments may vary with
time. Such processes are known as wide-sense stationary
processes.
We model and analyse telephone traffic in segments when they
can be considered to be stationary. In our modelling we use
discrete state stochastic processes.
27
Markov Processes
A.A. Markov proposed a simple and highly useful form of
dependency among the random variable forming a stochastic
process. A discrete time Markov chain, i.e discrete time
discrete state Markov process is defined
Equation says that, the duration for which a process has stayed
in a particular state does not influence the next state transition.
28
1 1 1 1 1 1
1 1
[{ ( ) }/ { ( ) , ( ) ,... ( ) }]
[{ ( ) }/ { ( ) }
n n n n n n
n n n n
P X t x X t x X t x X t x
P X t x X t x
+ +
+ +
= = = =
= = =
Markov Processes
There are only two distribution function that satisfy this
criterion.
One is the exponential distribution, which is continuous , and
The other is the geometric distribution which is discrete.
Thus, the interstate transition time in a discrete time Markov
Process is geometrically distributed and in a continuous time
Markov process, it is exponentially distributed.
In view of above described property of these distributions,
they are said to be memoryless.
29
Birth-Death Processes
If we apply the restriction that the state transition of a Markov
Chain can occur only to the adjacent states, then we can obtain
Birth- Death (B-D) Proceses.
The number in the population is a random variable and
represents the state value of the process.
The B-D process moves from its state k to state k-1 if a death
occurs or moves to state k+1 if a birth occurs, and stays in the
same state if there is no birth or death during the time period
under consideration, as shown in Fig. 8.4 .
30
Birth-Death Processes
31
At Time t + t
At Time t
k +1
k
k -1
k
Death
No Change
Birth
Fig. 8. 4. State transitions at a Birth Death process
Birth-Death Processes
A telecommunication network can be modelled as a B-D
process, where the number of busy servers represents the
population, a call request means a birth and call termination
means a death.
In order to analyse a B-D process, we consider a time interval
t small enough such that:
1) There can almost be only one state transition in that
interval.
2) There is only one arrival or one termination but not both,
and
3) There may be no arrival or termination leaving the state
unchanged in the time interval t.
32
Birth-Death Processes
We further assume that,
The probability of an arrival or termination in a particular
interval is independent of what had happened in the earlier
time intervals and.
The probability of an arrival is directly proportional to the time
interval t .
Let ,
P
k
(t) = The probability that the system is in the state k at time
t, i.e k servers are busy at time t.

k
= Call arrival rate in state k.

k
= Call termination rate in state k
33
Birth-Death Processes
Then, probabilities in the time interval t :
P[exactly one arrival] = t
P[exactly one termination] = t
P[no arrival] = 1 - t
P[no termination] = 1 - t
Probability of finding the system in state k at time t + t
P
k
(t + t )
= P
k -1
(t)
k-1
t + P
k +1
(t)
k+1
t + (1 -
k
t )(1 -
k
t )P
k
(t)
The first term on RHS represents the possibility of finding the
system in state k-1 at time t and a birth or a call request
occurring during the interval t to t+ t .
34
Birth-Death Processes
The possibility of finding system in state k+1 at time t and a
death or a call termination occurring during the interval t to t+
t is given by the second term.
The last term shows the no arrival and no termination case.
Expanding the equation and ignoring the second order t term
P
k
(t + t ) =
= P
k -1
(t)
k-1
t + P
k +1
(t)
k+1
t - (
k
+
k
)P
k
(t)t +P
k
(t)
Rearranging the terms
35
1 1 1 1
( ) ( )
( ) ( ) ( ) ( )
k k
k k k k k k k
P t t P t
P t P t P t
t

+ +
+ A
= + +
A
Birth-Death Processes
In the limit t0, we get
For k =0, i.e. no calls in progress, there can be no termination
of a call so
0
= 0. Further there can be no state with -1 as the
state value. So
Under Steady state condition
36
1 1 1 1
( )
( ) ( ) ( ) ( )
k
k k k k k k k
dP t
P t P t P t
dt

+ +
= + +
0
1 1 0 0
( )
( ) ( )
dP t
P t P t
dt
=
( )
0
k
dP t
dt
=
Birth-Death Processes
So B D process becomes stationary. Therefore, the steady
state equation of B-D process are

37
1 1 1 1
1 1 0 0
( ) 0 1
0 0
k k k k k k k
P P P for k
P P for k


+ +
+ + = >
= =
Incoming Traffic and Service Time Characterisation
When subscriber originates the call, he adds one to the number
of calls arriving at the network.
To model originating process, we define zero death rate. This
is also known as a renewal process.
It is a pure birth process in the sense that it can only add to the
population as the time goes by and cannot reduce the
population by itself.
So considering
k
= 0 in B-D process
38
1 1
0
0 0
( )
( ) ( ) 0
( )
( ) 0
k
k k k k
dP t
P t P t for k
dt
dP t
P t for k
dt

= >
= =
Incoming Traffic and Service Time Characterisation
As soon as a birth occurs at t = t
1
, it is impossible to find the
system ins state 0.
Poisson process is used to find the probability of k births in a
given time interval.
1)

Assuming at t =0, the system is in state zero, i.e. no births

have taken place.
39
1
0
0
( )
( ) ( ) 1
( )
( ) 0
k
k k
dP t
P t P t for k
dt
dP t
P t for k
dt

= >
= =
1 0
(0)
0 0
k
for k
P
for k
=

=
`
=
)
Incoming Traffic and Service Time Characterisation
With these condition we get the solution for probability of 0
arrival in the time interval t.
2)
For k =1
Solving this equation
For k = 2, the solution is
40
0
( )
t
P t e

=
1
1
( )
( )
t
dP t
P t e
dt

= +
1
( )
t
P t te

=
2
2
( )
( )
2!
t
t e
P t

=
Incoming Traffic and Service Time Characterisation
By induction, general solution
3)
Equation (3) is the Poisson arrival process equation. It
represents the probability of k arrival in the time interval t.
41
( )
( )
!
k t
k
t e
P t
k

=
Example 6
A rural telephone exchange normally experiences four call
origination per minute. What is the probability that exactly
eight calls occur in an arbitrarily chosen interval of 30 seconds
?
Answer
= 4/60 = 1/15 calls per seconds.
When t = 30s, t = 2
Then probability of exactly eight arrivals is given by
0.00086
42
8 2
8
(2)
( )
8!
e
P t

=
Example 7
A switching system serves 10000 subscribers with a traffic
intensity of 0.1 E per subscriber. If there is a sudden spurt in
the traffic ,increasing the average traffic by 50 %, what is the
effect on the arrival rate ?
Answer:
Number of Active subscribers during
A) Normal Traffic = 1000. B) increased traffic =1500
Number of available subscribers for generating new traffic
during
A) Normal Traffic = 9000 . B) increased traffic = 8500
Change in the arrival rate = (500/9000) X 100 = 5.6 %
43
Incoming Traffic and Service Time Characterisation
44
Fig. 8.5 Relationship among different Markov Process in the form of Venn diagram.
Poisson Process

k
=
Renewal Process

k
= 0
B-D Process

k
0

k
0
Incoming Traffic and Service Time Characterisation
Venn diagram, we may describe the Poisson process as
1) A pure birth process with constant birth rate.
2) A birth-death process with zero death rate and a constant
birth rate.
3) A Markov process with state transitions limited to the next
higher state or to the same state, and having a constant
transition rate.
45
Incoming Traffic and Service Time Characterisation
Example of Poisson process
1) Number of telephone calls arriving at an exchange.
2) Number of coughs generated in a medical ward by the
patients
3) Number of rainy days in a year.
4) Number of typing errors in a manuscript.
5) Number of bit errors occurring in a data communication
system.
46
Incoming Traffic and Service Time Characterisation
Let us consider a pure death process.
Considering pure death process
47
1
0
1
( )
( ) ( ) 1
( )
( ) 0
k
k k
dP t
P t P t for k
dt
dP t
P t for k
dt

+
= >
= =
1
0
1
( )
( ) ( ) 1
( )
( ) 0
k
k k
dP t
P t P t for k
dt
dP t
P t for k
dt

+
= >
= =
Incoming Traffic and Service Time Characterisation
Assuming t = 0 and assuming suitable boundary condition
Solving these equations,
4)
48
1
0
1
( )
( ) ( ) 0
( )
( )
( )
( ) 0
k
k k
N
N
dP t
P t P t for k N
dt
dP t
P t for k N
dt
dP t
P t for k
dt

+
= < <
= =
= =
1
0
( )
( )
( ) 0
( )!
( )
( ) 0
( 1)!
t
N
N k
t
k
N
t
P t e for k N
t
P t e for k N
N k
t
P t e for k
N

= =
= < <

= =

Incoming Traffic and Service Time Characterisation

Equation (4) expresses the probability of no termination or
death in a given interval as the initial level. It is the probability
distribution of the service times or the holding times in the
case of calls in a switching system.
The holding times of conventional voice conversations are
well approximated by an exponential distribution. Let t
m
be the
constant holding time. Then, the probability of finding k busy
channels at any instant of time is merely the probability that
there are k arrivals during the time interval t
m
immediately
preceding the instant of observation
49
( ) exp( )
( )
!
k
m m
k m
t t
P t
k

=
Blocking Models and Loss Estimates
The behavior of a loss system is studied by using blocking
models and that of the delay systems by using queuing models.
Three aspects of telecommunication system
1) Modelling the system : B-D process
2) Traffic arrival Model : Poisson Process
3) Service time distribution: Considering for holding time as
an exponential or constant time distribution.
50
Blocking Models and Loss Estimates
There are three ways in which overflow traffic may be
handled:
1) Lost Call Cleared (LCC) : The traffic rejected by one set of
resources may be cleared by another set of resources in the
network.
2) Lost Calls Returned (LCR) : The traffic may return to the
same resources after sometime.
3) Lost Calls Held (LCH) : The traffic may be held by the
resources as if being serviced but actually serviced only after
the resources become available.
51