Basic teletraffic concepts

An intuitive approach
(theory will come next)

Focus on “calls”

Giuseppe Bianchi
 1 user making phone calls
TRAFFIC is a “stochastic process”

BUSY 1

IDLE 0
time
 How to characterize this process?
 statistical distribution of the “BUSY” period
 statistical distribution of the “IDLE” period
 statistical characterization of the process “memory”
 E.g. at a given time, does the probability that a user starts a call result
different depending on what happened in the past?

Giuseppe Bianchi
 Traffic characterization
suitable for traffic engineering
amount of busy time in t
traffic intensity A i  lim 
t  t
  average number  of calls per min    average call duration  min  
 probability that, at a random time t, user is in BUSY state 
 mean process value

All equivalent (if stationary process)

Giuseppe Bianchi
 Traffic Intensity: example

User makes in average 1 call every

hour
Each call lasts in average 120 s
Traffic intensity =
120 sec / 3600 sec = 2 min / 60 min = 1/30
Probability that a user is busy
User busy 2 min out of 60 = 1/30

adimensional
Giuseppe Bianchi
 Traffic generated by more than
one users
U1 Traffic intensity
(adimensional, measured in Erlangs):
4
U2
A   Ai  4 Ai
i 1
U3

U4

 4 k
 
P k active calls    Ai 1  Ai 
4k

k
E  active calls  4  Ai  A
TOT
Giuseppe Bianchi
 example
 5 users
 Each user makes an average of 3 calls per
hour
 Each call, in average, lasts for 4 minutes
 calls  4 1
Ai  3    hours    erl 
 hour  60 5
1
A  5   erl   1 erl 
5 number of active users
0
probability
0,327680
Meaning: in average, there is 1 active call; 1 0,409600
but the actual number of active calls varies 2
3
0,204800
0,051200
from 0 (no active user) to 5 (all users active), 4 0,006400
with given probability 5 0,000320

Giuseppe Bianchi
 Second example
n. active users binom probab cumulat
0 1 1,3E-01 0,126213
1 30 2,7E-01 0,396669
2 435 2,8E-01 0,676784
3 4060 1,9E-01 0,863527
 30 users 4 27405 9,0E-02 0,953564
 Each user makes an 5
6
142506
593775
3,3E-02
1,0E-02
0,987006
0,996960
average of 1 calls per 7 2035800 2,4E-03 0,999397
hour 8
9
5852925
14307150
5,0E-04
8,7E-05
0,999898
0,999985
 Each call, in average, 10 30045015 1,3E-05 0,999998
11 54627300 1,7E-06 1,000000
lasts for 4 minutes 12 86493225 1,9E-07 1,000000
13 119759850 1,9E-08 1,000000
 4 14 145422675 1,7E-09 1,000000
A  30  1    2 Erlangs 15 155117520 1,3E-10 1,000000
 60  16
17
145422675
119759850
8,4E-12
5,0E-13
1,000000
1,000000
SOME NOTES: 18 86493225 2,6E-14 1,000000
19 54627300 1,2E-15 1,000000
-In average, 2 active calls (intensity A); 20 30045015 4,5E-17 1,000000
-Frequently, we find up to 4 or 5 calls; 21
22
14307150
5852925
1,5E-18
4,5E-20
1,000000
1,000000
-Prob(n.calls>8) = 0.01% 23 2035800 1,1E-21 1,000000
24 593775 2,3E-23 1,000000
-More than 11 calls only once over 1M 25 142506 4,0E-25 1,000000
26 27405 5,5E-27 1,000000
27 4060 5,8E-29 1,000000
TRAFFIC ENGINEERING: how many 28 435 4,4E-31 1,000000
29 30 2,2E-33 1,000000
channels to reserve for these users! 30 1 5,2E-36 1,000000

Giuseppe Bianchi
 A note on binomial coefficient computation
 60  60!
    1.39936e  12
12  12!48!
but 60! 8.32099e  81 (overflow problems!! )
 60    60  
   exp log    exp log 60!  log12!  log 48!  
12   12  
 60 12 48

 exp  log i    log i    log i   (no overflow!! before exp...)
 i 1 i 1 i 1 

 60  12
  Ai 1  Ai  48 
12 
 60 12 48

 exp  log i    log i    log i   12 log Ai   48 log1  Ai  
 i 1 i 1 i 1 
(no overflow! ! never! )

Giuseppe Bianchi
 Infinite Users
Assume M users, generating an overall traffic intensity A
(i.e. each user generates traffic at intensity Ai =A/M).
M
We have just found that  A
k  1  
M  M!  A   M 
P k active calls, M users    Aik 1  Ai 
M k
  
k   M  k !k!  M   A  k
1  
 M
Let Minfinity, while maintaining the same overall traffic intensity A
k M k
M! 1 A  A  A
P k active calls,  users   lim    1    1   
M   M  k ! k! M k
 M  M
A
 M

M  M  1   M  k  1 
 k
Ak A A
  A Ak
  lim  1    1    e A
k! M   M k
 M    M k!
 

Giuseppe Bianchi
 Poisson Distribution
30%
poisson
A=2 erl
25% binomial (M=30)

20%
A=10 erl
15%

10%

5%

0%
0 2 4 6 8 10 12 14 16 18 20 22

k Very good matching with Binomial

A
Pk  A  e A (when M large with respect to A)
k! Much simpler to use than Binomial
(no annoying queueing theory complications)
Giuseppe Bianchi
 Limited number of channels
THE most important problemU1
in circuit switching
 The number of channels
C is less than the number U2
of users M (eventually
infinite) X
 Some offered calls will be
U3
“blocked”
 What is the blocking U4 X
probability?
 We have an expression for
P[k offered calls]
 We must find an expression for
P[k accepted calls]
 As: TOT
P[block]  P C accepted calls No. carried calls versus t
No. offered calls versus t
Giuseppe Bianchi
 Channel utilization probability
offered traffic: 2 erl - C=3

 C channels available 35%

 Assumptions: 30%

 Poisson distribution (infin. users) 25% offered calls

 Blocked calls cleared 20% accepted calls
 It can be proven (from
15%
Queueing theory) that:
10%
P[k calls in the system, k  (0, C)] 
5%
P k offered calls
 C 0%

 P i offered calls
i 0
0 1 2 3 4 5 6 7 8

(very simple result!)

P C offered calls
 Hence: P[system full]  P[C accepted calls]  C

 P i offered calls
i 0
Giuseppe Bianchi
 Blocking probability: Erlang-B
 Fundamental formula for  Efficient recursive computation
telephone networks planning available
 Ao=offered traffic in Erlangs AE A  o
E1,C  Ao   o 1,C 1

C  Ao E1,C 1  Ao 
AoC 100,00%

 block  C C! j  E1,C  Ao  blocking probability

Ao 10,00%

j  0 j! 1,00%
C=1,2,3,4,5,6,7
0,10%

0,01%
0 1 2 3 4 5
offered load (erlangs)

Giuseppe Bianchi
 NOTE: finite users
 Erlang-B obtained for the  Erlang-B can be re-obtained as
infinite users case limit case
 It is easy (from queueing  Minfinity
theory) to obtain an explicit  Ai0
blocking formula for the
 M·AiAo
finite users case:

 ENGSET FORMULA:  Erlang-B is a very good

approximation as long as:
 M  1
A 
i
C
  A/M small (e.g. <0.2)
 C 
 block  C  In any case, Erlang-B is a
k  M  1
 Ai   conservative formula
k 0  i   yields higher blocking
probability
Ao  Good feature for planning
Ai 
M
Giuseppe Bianchi
 Capacity planning
 Target: support users with a given Grade Of
Service (GOS)
 GOS expressed in terms of upper-bound for the blocking probability
GOS example: subscribers should find a line available in the 99% of
the cases, i.e. they should be blocked in no more than 1% of the
attempts
 Given:
C channels
Offered load Ao
Target GOS Btarget
 C obtained from numerical inversion of

Btarget  E1,C  Ao 

Giuseppe Bianchi
 Channel usage efficiency
Offered load (erl) Carried load (erl)

Ao C channels Ac  Ao 1  B 

Ao B

Blocked traffic
Ac Ao 1  E1,C  Ao   Ao
efficiency :     if small blocking
C C C
Fundamental property: for same GOS, efficiency
increases as C grows!! (trunking gain)
Giuseppe Bianchi
 example
100,0%
A= 40 erl
A= 60 erl
A= 80 erl
blocking probability

A= 100 erl
10,0%

1,0%

0,1%
0 20 40 60 80 100 120
capacity C

GOS = 1% maximum blocking. 40 erl C >= 53  = 74.9%

60 erl C >= 75  = 79.3%
Resulting system dimensioning 80 erl C >= 96  = 82.6%
and efficiency: 100 erl C >= 117  = 84.6%
Giuseppe Bianchi
Erlang B calculation - tables

Giuseppe Bianchi