TRAFFIC ANALYSIS

Arrival Distributions
The most fundamental assumption of classical traffic analysis is
that call arrivals are independent.
An arrival from one source is unrelated to an arrival from any
other source.
Even though this assumption may be invalid in some instances,
it has general usefulness for most applications.
In those cases where call arrivals tend to be correlated, useful
results can still be obtained by modifying a random arrival
analysis.
In this manner the random arrival assumption provides a
mathematical formulation that can be adjusted to produce
approximate solutions to problems that are otherwise
mathematically intractable.
Negative Exponential Inter-arrival Times
The probability distribution of inter-arrival times is



The equation defines the probability that no
arrivals occur in a randomly selected interval t.
This is identical to the probability that t sec
elapse from one arrival to the next.
0
( )
t
P t e

=
Example
Assuming each of 10,000 subscriber lines originate one call
per hour, how often do two calls arrive with less than 0.01 sec
between them?

Solution
The average arrival rate
=(10,000)(1/3600) = 2.78 arrivals per second
the probability of no arrival in a 0.01-sec interval is


Thus 2.7% of the arrivals occur within 0.01 sec of the previous
arrival.
Since the arrival rate is 2.78 arrivals per second, the rate of
occurrence of inter-arrival times less than 0.01 sec is
(2.78)(0.027) = 0.075 times a second
0.0278
0
(0.0278) 0.973 P e

= =
Negative Exponential Inter-arrival Times
To designate the average call arrival rate from a large
group of independent sources (subscriber lines) as
use the following assumptions:
Only one arrival can occur in any sufficiently small interval.
The probability of an arrival in any sufficiently small interval
is directly proportional to the length of the interval. (The
probability of an arrival is At, where At is the interval
length.)
The probability of an arrival in any particular interval is
independent of what has occurred in other intervals.
Negative Exponential Interarrival Times
The first two assumptions made in deriving the negative
exponential arrival distribution, can be intuitively justified for
most applications.
The third assumption implies certain aspects of the sources that
cannot always be supported.
First, certain events, such as television commercial breaks
might stimulate the sources to place their calls at nearly the
same time. In this case the negative exponential distribution
may still hold, but for a much higher calling rate during the
commercial.
Negative Exponential Interarrival Times
A more subtle implication of the independent arrival assumption
involves the number of sources, and not just their calling
patterns.
When the probability of an arrival in any small time interval is
independent of other arrivals, it implies that the number of
sources available to generate requests is constant.
If a number of arrivals occur immediately before any subinterval
in question, some of the sources become busy and cannot
generate requests.
The effect of busy sources is to reduce the average arrival rate.
Thus the inter-arrival times are always somewhat larger than
what the equation predicts them to be.
The only time the arrival rate is truly independent of source
activity is when an infinite number of sources exist.

Negative Exponential Interarrival Times
If the number of sources is large and their average activity is
relatively low, busy sources do not appreciably reduce the
arrival rate.
For example, consider an end office that services 10.000
subscribers with 0.1 erlangs of activity each.
Normally, there are 1000 active links and 9000 subscribers
available to generate new arrivals.
If the number of active subscribers increases by an unlikely
50% to 1500 active lines, the number of idle subscribers
reduces to 8500, a change of only 5.6%.
Thus the arrival rate is relatively constant over a wide range of
source activity.
Whenever the arrival rate is fairly constant for the entire range
of normal source activity, an infinite source assumption is
justified.
Poisson Arrival Distribution
Exponential equation merely provides a means of determining
the distribution of inter-arrival times.
It does not, by itself, provide the generally more desirable
information of how many arrivals can be expected to occur in
some arbitrary time interval.
Using the same assumptions presented, however, the
probability of arrivals in an interval t can be determined as



The equation is the well-known Poisson probability law.
when j = 0, the probability of a no arrivals in an interval t is P
0
(t),
as obtained in exponential equation.
( )
( )
!
j
t
j
t
P t e
j

=
Poisson Arrival Distribution
Poisson equation assumes arrivals are independent and occur
at a given average rate A, irrespective of the number of arrivals
occurring just prior to an interval in question.
Thus the Poisson probability distribution should only be used for
arrivals from a large number of independent sources.
The equation defines the probability of experiencing exactly j
arrivals in t sec.
Usually there is more interest in determining the probability of j
or more arrivals in t sec:




where P
i
(t) is defined in poisson equation

1
0
( ) ( ) 1 ( ) 1 ( )
j
j i i j
i j i
P t P t P t P t

> <
= =
= = =

Example
Given a message-switching node that normally experiences four
arrivals per minute, what is the probability that eight or more
arrivals occur in an arbitrarily chosen 30-sec interval?

Solution

The average number of arrivals in a 30-sec interval is


The probability of eight or more arrivals (when the average is 2) is


30
(4) 2
60
t
| |
= =
|
\ .
( )
8
8
7
0
1 2 3 4 7
2
(2) (2)
1 2
2 2 2 2 2
1 (1 ... )
1! 2! 3! 4! 7!
0.0011
i
i
i
i
P P
P
e

>
=
=

=
=
= + + + + + +
=

Example
What is the probability that a 1000 bit data block
experiences exactly four errors while being transmitted
over a transmission link with a bit error rate (BER) = 10
-5
?
Solution
Assuming independent errors (a questionable assumption
on many transmission links), we can obtain the probability
of exactly four errors directly from, the Poisson
distribution. The average number of errors (arrivals)
t = (10
3
)10
-5
= 0.01.
Thus,
prob(4 errors) = P
4
(0.01) = [(0.01)
4
/4!]e
-0.01
=4.125(10)
-10

Holding Time Distributions
The second factor of traffic intensity is the average
holding time t
m
.
In some cases the average of the holding times is all
that needs to be known about holding times to
determine blocking probabilities in a loss system or
delays in a delay system.
In other cases it is necessary to know the probability
distribution of the holding times to obtain the desired
results.
The two most commonly assumed holding time
distributions are:
constant holding times
exponential holding times.
Constant Holding Times
Constant holding times cannot be assumed for
conventional voice conversations
It is a reasonable assumption for such activities as
per-call call processing requirements, interoffice
address signaling, operator assistance, and recorded
message playback.
Constant holding times are obviously valid for
transmission times in fixed length packet networks.
When constant holding time messages are in effect,
it is straightforward to use Poison equation to
determine the probability distribution of active
channels.
Constant Holding Times
Assume, for the time being, that all requests are
serviced, then the probability of j channels being busy
at any particular time is merely the probability that j
arrivals occurred in the time interval of length t
m

immediately preceding the instant in question.
Since the average number of active circuits over all
time is the traffic intensity A = t
m
, the probability of j
circuits being busy is dependent only on the traffic
intensity:



where = the arrival rate
t
m
= the constant holding time
A = the traffic intensity in erlangs

( ) ( )
!
j
A
j m j
A
P t P A e
j

= =
Exponential Holding Times
The most commonly assumed holding time distribution for
conventional telephone conversations is the exponential holding
time distribution:





where t
m
is the average holding time.
The equation specifies the probability that a holding time
exceeds the value t.
This relationship can be derived from a few simple assumptions
concerning the nature of the call termination process.
Its basic justification lies in the fact that observations of actual
voice conversations exhibit a remarkably close correspondence
to an exponential distribution.
/
( )
m
t t
P t e

> =
Exponential Holding Times
The exponential distribution possesses the curious property that
the probability of a termination is independent of how long a call
has been in progress.
No matter how long a call has been in existence, the probability
of it lasting another t sec is defined by the equation
In this sense exponential holding times represent the most
random process possible.
Not even knowledge of how long a call has been in progress
provides any information as to when the call will terminate.
Exponential Holding Times
Combining a Poisson arrival process with an exponential holding
time process to obtain the probability distribution of active circuits
is more complicated than it was for constant holding times
because calls can last indefinitely.
The result proves to be dependent on only the average holding
time.
Thus the constant holding time equation is valid for exponential
holding times also (or any holding time distribution).
The probability of j circuits being busy at any particular instant,
assuming a Poisson arrival process and that all requests are
serviced immediately is


where A is the traffic intensity in erlangs.
This result is true for any distribution of holding times.
( )
!
j
A
j
A
P A e
j

=
Example
Assume that a trunk group has enough channels to immediately carryall
of the traffic offered to it by a Poisson process with an arrival rate of one
call per minute. Assume that the average holding time is 2 min. What
percentage of the total traffic is carried by the first five circuits, and how
much traffic is carried by all remaining circuits? (Assume that the traffic
is always packed into the lowest-numbered circuits.)

Solution
The traffic intensity (offered load) of the system is A=(l)(2) = 2 erlangs. The
traffic intensity carried by i active circuits is exactly i erlangs.
Hence the traffic carried by the first five circuits can be determined as
follows:






All of the remaining circuits carry (2 - 1.89) = 0.11 erlangs
5 1 2 3 4 5
2 3 4 5
-2
A = 1 P (2) + 2P (2) + 3P (2) + 4P (2) + 5P (2)
2(2) 3(2) 4(2) 5(2)
=e 2
2! 3! 4! 5!
1.89erlangs
(
+ + + +
(

=
Example
The result of the Example demonstrates the
principle of diminishing returns as the
capacity of a system is increased to carry
greater and greater percentages of the
offered traffic.
The first five circuits in the example carry
94.5% of the traffic while all remaining
circuits carry only 5.5% of the traffic.
If there are 100 sources. 95 extra circuits are
needed to carry the 5.5%.