Teletraffic Engineering

Dr. Hicham Aroudaki Damascus, 17th October 2009

Traffic Engineering Purpose p

Traffic theory is used to perform cost-effective dimensioning of network infrastructure equipment.

Traffic Engineering
Interesting g Questions

Given the system and incoming traffic, what is the quality of service experienced by the user? Given the incoming traffic and required quality of service, how should the system be dimensioned? Gi Given the th system t and d required i d quality lit of f service, what is the maximum traffic load?
2

Traffic Engineering
Interesting g Questions Q Qualitatively, y, the relationships p are as follows:

To describe the relationships quantitatively, mathematical models are needed.

3

Traffic Engineering
Disciplines p & Goals Traffic theory y is based on the following disciplines:

Practical g goals:

probability theory stochastic t h ti processes queueing theory statistical analysis (analysis of measurement data) operations analysis optimization theory decision analysis (Markov decision processes) simulation techniques

Network planning

Di Dimensioning i i Optimization performance analysis

Network management and control

efficient operating fault recovery traffic management routing ti

4

Demand for terms additional capacity Defining Problem the initial approach Offeredof vs. carried traffic Offered traffic: traffic as it is originally generated in the sources Carried traffic: traffic as it is carried by the network

Network

offered traffic

carried traffic blocked traffic

Demand for additional capacity Problem of the initial approach Characterization of carried Circuit-switched traffic

traffic

number of ongoing calls or active connections (Erl) may be converted into bit rate in digital systems (e.g. a telephone p call reserves 64 kbps p ( (= 8000*8 bps) p ) in a PCM system)

Packet-switched traffic
bit stream (bps, kbps, Mbps, Gbps, ) packet stream (pps) number of active flows (Erl)

Defining terms
Busy y Hour
It is the given period within a d th day that tb bears th the hi highest h t traffic intensity. This period usually has the highest amount of blocked or lost calls. The 'busy hour' traffic is used to work out the equipment quantities of the network.

If the dimensioning of equipment at this period is correct and blocked calls can be minimized, all other nonbusy hour traffic should then be handled satisfactorily.

110 100 90 80 70 60 50 40 30 20 10
0 3 6

Busy Hour

No. of Calls

12

15

18

21

24

Daily y traffic profile p measured in Syria y

9% 8% 8% 8% 7% 6% 6% 5% 5% 4% 3% 3% 2% 2% 1% 0% 0:00 1:00 2:00 3:00 4:00 5:00 6:00 7:00 8:00 9:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00 18:00 19:00 20:00 21:00 22:00 23:00 1% 1% 0% 0% 1% 1% 2% 3% 4% 6% 6% 6% 6% 6%

# of Minutes
7% Percenta age of total traff fic 6%

7%

# of Calls

4%

HouroftheDay

Defining Terms
Blocking g
In a loss system some calls are lost a call is lost if all n channels are occupied when the call arrives the term blocking refers to this event There are two different types yp of blocking gq quantities: Call blocking Bc = probability that an arriving call finds all n channels occupied = the fraction of calls that are lost Time blocking Bt = probability that all n channels are occupied at an arbitrary ti time = th the f fraction ti of f time ti that th t all ll n channels h l are occupied i d The two blocking quantities are not necessarily equal Example: your own mobile If calls arrive according to a Poisson process, then Bc = Bt Call blocking is a better measure for the quality of service experienced by the subscribers but, typically, time blocking is easier to calculate

Defining Terms
Grade of Service (GoS) ( )
Is a measure of the call blocking (the ability to make call during the busiest time). It is typically given as the likelihood that a call is blocked or the likelihood of a call experiencing a delay greater than a certain queuing time. Is determined by the available number of channels and used to estimate the total pp number of users that a network can support. In general, GOSis measured by looking at traffic carried, traffic offered, and calculating the traffic blocked and lost. The proportion of lost calls is the measure of GOS. GOS For cellular circuit groups an acceptable GoS = 0.02. This means that two users of the circuit group out of a hundred will encounter a call refusal during the busy hour at the end of the planning period.
GOS
= traffic lost / traffic offered = proportion of time for which congestion exists = probability b bilit of f congestion ti or bl blocking ki probability b bilit = probability that a call will be lost due to congestion

Defining Terms
Quality y of Service
The standard metrics used to measure the QoS:

Coverage: the strength of the measured signal is used to estimate the size of the cell cell.

Accessibility (includes Grade of Service): is about determining the ability of the network to handle successful calls from mobile-to-fixed networks and from mobile-to-mobile networks.

Audio quality: monitoring a successful call for a period of time for the clarity of the communication channel.

Simple p ( (Generic) ) teletraffic model

Customers arrive at rate (customers per time unit) 1/ / = average inter-arrival time Customers are served by n parallel servers When busy busy, a server serves at rate (customers per time unit) 1/ = average service time of a customer There are n + m customer places in the system at t least l t n service i places l and d at t most t m waiting iti places l It is assumed that blocked customers (arriving in a full system) are lost

12

Model Classification
Pure loss system y
Finite number of servers (n < ), n service places, no waiting places (m = 0) If the system is full (with all n servers occupied) when a customer arrives, it is not served at all but lost From the customers point of view, it is interesting to know e.g. What is the probability that the system is full when it arrives?

13

Model Classification
Infinite System y
Infinite number of servers (n = ), no waiting places (m = 0) No customers are lost or even have to wait before getting served Sometimes, this hypothetical yp model can be used to g get some approximate pp results for a real system (with finite system capacity) Always, it gives bounds for the performance of a real system (with finite system capacity) it is much easier to analyze than the corresponding finite capacity models

14

Model Classification
Pure q queuing g system y
Finite number of servers (n < ), n service places, infinite number of waiting places (m = ) If all n servers are occupied when a customer arrives, it occupies one of the waiting places No customers are lost but some of them have to wait before getting served From the customers point of view, it is interesting to know e.g. what is the probability that it has to wait too long?

15

Model Classification
Lossy yq queuing g system y
Finite number of servers (n < ), n service places, finite number of waiting places (m < ) If all n servers are occupied but there are free waiting places when a customer arrives, it occupies one of the waiting places If all n servers and all m waiting places are occupied when a customer arrives, it is not served at all but lost Some customers are lost and some customers have to wait before getting served

16

Towards detailed modeling

Telephone p traffic model
Telephone traffic consists of calls

a call occupies one channel from each of the links along its route call characterization: holding time (in time units)

Modeling g of offered traffic

call arrival process (at which moments new calls arrive) holding time distribution (how long they take)

17

Towards detailed modeling

Telephone p traffic model
Link model: a pure loss system

a server corresponds p to a channel the service rate depends on the average holding time the number of servers, n, depends on the link capacity when all channels are occupied, call admission control rejects new calls so that they will be blocked and lost

Modeling of carried traffic

traffic process tells the number of ongoing calls = the number of occupied channels

18

Towards detailed modeling

Traffic process p

Traffic intensity is the average number of calls simultaneously in progress during a particular period of time.
19

Traffic Measurement Unites

Agner Krarup Erlang was born in 1878 in Lnborg, Denmark. Through his studies of telecommunications traffic, he proposed a formula to calculate the fraction of callers served by a village exchange who would have to wait when attempting to place a call to someone outside the village village. In 1909, he published his first work: The Theory of Probabilities and Telephone Conversations. He gained worldwide recognition for his work, and his formula was accepted for use by the General Post Office in the UK. He worked for the Copenhagen Telephone Company for twenty years, until his death in 1929. During the 1940s, the Erlang became the accepted unit of telecommunication traffic measurement.

A.K. Erlang, 1878-1929

Traffic Measurement Unites An Intuitive Definition

Traffic or traffic intensity is a non-physical measure of load on a system. It is thus given by a pure number with no physical unit attached to it.
The load is simply a zero/one matter of a server being free/occupied. A server may be any type of resource entity that has this dual property (line, (line trunk, trunk switch inlet or outlet, signal receiver, radio channel, memory access, etc.). It has been decided to use the notation Erlang as a traffic unit. Thus a single server carries a traffic of 1 Erlang g if it is continuously y occupied p during g an observation p period. Two servers with occupations 1/4 and 3/4 of the time also together carry 1 Erlang. Traffic is normally related to a traffic carrying system, consisting of a discrete number of servers. Each of the servers can at any moment carry a load of one or zero. A system of n servers can carry an instantaneous i t t l d A of load f any integer i t number b 0 A n. The definition implies that two servers of different capacity (say one line of 9.6 kb/s and one of 64 kb/s) both carry 1 Erlang as long as they are occupied to their full capacity, even though the amounts of data transmitted during the same time interval are very different.

21

Traffic Measurement Unites Basic Definition

t1 t2 t3 t4 t
4

At any point i t of f time ti the th resource (e.g. ( channel) h l) is i loaded l d d or not. t During the observation time the occupation (load) time is:

T tot = t i

Percentage of occupation during the observation time is: T tot By definition, Traffic Intensity (I): Measuring unit: Erlang (Erl)
1 Erlang:

/T

i =1

I = T tot / T

1 hour of continuous use of one channel = 1 Erlang 1 Erlang = 1 hour (60 minutes) of traffic In data communications, an 1 E = 64 kbps of data
22

Traffic Intensity calculation Littels formula

t1 t2 t3 t4 t

Traffic Intensity: I =

t
i =1

N ct = = nc t T

Traffic Intensity is the product of f the th call ll arrival i l rate t and d th the mean duration of calls handled by the channel (mean holding time)

23

Example p
If a group of users made 30 calls in one hour, hour and each call had an average call duration of 5 minutes, what is the corresponding Erlang value ?

Example p
If a group of users made 30 calls in one hour, hour and each call had an average call duration of 5 minutes, what is the corresponding Erlang value ?
Traffic Intensity = 30 x 5 / 60 = 2.5 Erlangs

Example p
Consider the p pattern of activity y in a cell of capacity p y 10 channel over a period of 1 hour. The rate of calls p per minute is 97/60. The average holding time per call, in minutes is 294/97. How many channels are occupied in average?

26

Example p
Consider the p pattern of activity y in a cell of capacity p y 10 channel over a period of 1 hour. The rate of calls p per minute is 97/60. The average holding time per call, in minutes is 294/97. How many channels are occupied in average?

I =(97/60)(294/97) = 4.9 Erlangs. g 4.9 channels are engaged. g g That is, on average,

27

Example p I
A call was established at 1am between a mobile and MSC. Assuming a continuous connection and data transfer rate at 30 kbit/s, determine the traffic intensity if the call is terminated at 1:50 am.

Example p I
A call was established at 1am between a mobile and MSC. Assuming a continuous connection and data transfer rate at 30 kbit/s, determine the traffic intensity if the call is terminated at 1:50 am. Traffic intensity = (1 call)*(50 mins)*(1 hour/60 min) = 0.833 Erlang

Note, traffic intensity has nothing to do with the data rate, only the holding time is taken into account.

Erlangs g Formula
The Erlang B formula is expressed as probability that an arriving call finds all N channels (servers of the system) occupied. The assumptions in the Erlang B formula are: Traffic originates from an infinite number of traffic sources independently. Lost calls are cleared assuming a zero holding time. Number of trunks or service channels is limited. Inter-arrival times of call requests are independent of each other. The probability of a user occupying a channel (called service time) is based on an exponential distribution. Traffic requests (with rate ) are represented by a Poisson distribution implying exponentially distributed call interarrival times.

B A

Also called: Erlangs B-formula Erlangs blocking formula Erlangs loss formula Erlangs first formula
30

Erlang-B g Traffic Table

Servers (Channels) GoS Offered Traffic

31

Graphs p for Erlangs g blocking g function

Blo ocking Probabilit ty

Bloc cking Probability y

Number of channels

Number of channels

Offered traffic intensity

offered traffic intensity

Usage g of Erlangs g formula

Blocking Probability 0.01

Minimum number of needed channels n 5

Offered traffic 0.8 Erlang

33

Usage of Erlangs formula p y vs. traffic Capacity

Given the quality of service requirement that B < 1%, the required capacity N depends on the traffic A intensity as follows:

N(A) = min{i =1,2, . . . | Erl(i,A) < 0.01}

Cap pacity (N)

Traffic (A)
34

Usage of Erlangs formula y of service vs. traffic Quality

Given the capacity N= 20 channels, the required quality of service (1 B) depends on the traffic intensity A as follows:

1-B(A) = 1-Erl (20,A)

Qo oS (1-B)

Traffic (A)
35

Usage of Erlangs formula y of service vs. capacity p y Quality

Given the traffic intensity A= 15 Erlang, the required quality of service (1 B) depends on the capacity N as follows:

1-B(N) = 1-Erl (N,15)

Qo oS (1-B)

Capacity (N)
36

Example p
A single g GSM service p provider support pp 10 digital g speech p channels. Assume the probability of blocking is 1.0%. From the Erlang B table find the traffic intensity. How many 3 minutes of calls does this represent?

Erlang-B g Traffic Table

38

Example p
A single g GSM service p provider support pp 10 digital g speech p channels. Assume the probability of blocking is 1.0%. From the Erlang B table find the traffic intensity. How many 3 minutes of calls does this represent? F From the th Erlang El B Chart Ch t th the traffic t ffi intensity i t it = 4.5 4 5 Erlangs El
I = nc t

nc = 4.5 / (3 mins/60) = 90 calls

Traffic Intensity y Models

Erlang B Formula:

All blocked calls are cleared

Extended Erlang B:

Similar to Erlang B, but takes into account that a percentage of calls are immediately represented to the system if they encounter blocking (a busy signal) The retry percentage can be specified signal). specified.

Erlang C Formula:

Blocked oc ed ca calls s de delayed ayed o or held e d in queue indefinitely. de te y

Poisson Formula:

Blocked calls held in queue for a limited time only.

Binomial Formula:

Lost calls held

Poissons formula
The Poisson formula is used for designing trunks on a route for a given GoS. It is used in the United States. The assumptions in Poissons formula are: Traffic originates from an infinite number of independent sources Traffic density per traffic source is equal Lost calls are held. A limited number of trunks or service channels exist.

41

Erlangs g C Formula
The Erlang C formula assumes that a queue is formed to hold all requested calls that cannot be served immediately. Customers who find all N servers busy join a queue and wait as long as necessary to receive service. This means that the blocked customers are delayed. No server remains idle if a customer is waiting. The assumptions in the Erlang C formula are:

Traffic originates from an infinite number of traffic sources independently. Lost calls are delayed. Number of trunks or service channels is limited. The probability of a user occupying a channel (called service time) is based on an exponential p distribution. Calls are served in the order of arrival.

42

Exponential p distribution
Exponential distributions are a class of continuous probability distributions. Main parameter is: (rate) used to model the time between independent events that happen at a constant average g rate. Usage: If events are assumed to occur randomly in time (i.e. follow a Poisson process) and the average time between events equals q , then the time between each consecutive event will be distributed according to an exponential distribution. For example, p , if an insurer sees that some particular type of natural disaster occurs on average once every 5.5 years, the time between such consecutive disasters can be modeled as f (1/5.5) (1/5 5) years. years Memory less property: the time until the next event also follows an Exponential distribution.
PDF of Exponential Distribution

Mean: -1 Variance: -2

f (x ; ) = e x

43

Exponential p distribution

F (x ; ) = 1- e x

CDF of the Exponential Distribution

44

Real example p for an arrival p process

Measured distribution of arrivals in a subscriber group, matching to an exponential curve.

Myskja, A, Walmann, O O. A statistical study of telephone traffic data with emphasis on subscriber behavior. I: 7th international teletraffic congress, ITC 7. Stockholm 1973.

45

Poisson Distribution
Poisson distribution is a discrete probability distribution. It expresses the probability of a number of events occurring in a fixed period of time if these events occur with a known average rate rate, and are independent of the time since the last event. The probability that there are exactly k occurrences (k being a non-negative integer, k = 0, 1, 2, ...) is:

f (k ; ) = e

k!
Mean: Variance:

f (k ; ) = e

k
Probability Mass Function (PDF) For instance, if the events occur on average every 4 min, and you are interested in the number of events occurring in a 10 minute interval, you would use as model a Poisson distribution with = 10/4 = 2.5. 46

k!

is a positive real number, equal to the expected number of occurrences that occur during the given interval

Poisson distribution
The Poisson (t) distribution models

the number of occurrences of an event in a time t with an expected rate of l events per period t when the time between successive events follows a Poisson process.

f (k ; ) = e

k!
Mean: Variance:

Examples If is the mean time between events, as used by the E Exponential ti l di distribution, t ib ti th then = 1/ . For example, imagine that records show that a computer crashes on average once every 250 hours h of f operation ti ( =250 250 hours), then the rate of crashing is 1/250 crashes per hour. Thus a Poisson (1000/250) = P i Poisson(4) (4) distribution di t ib ti models d l the number of crashes that could occur in the next 1000 hours of operation.

Probability Mass Function (PDF)

47

Poisson ( (in other words) )

In a Poisson process with rate , the number of points occurring in a fixed length t has the Poisson distribution, or equivalently, the lengths of the intervals separating successive points are independent and have identical, exponential distributions
Service times

t1
Ch 1

t2

t3

t4

t5
Ch 3

t5

t6

t7

t8

t9

Arrival times Departure times

Profile of Typical yp Cellular usage g ( (1994) )

Capacity p y considerations ( (1) )

The capacity required in a certain service area is proportional to the traffic which can be served at a given quality of service (QoS) within this service area. The traffic generated by the subscribers within the service area is proportional to mean service time and the mean service requesting rate. Thus for circuit switched services the traffic is given by:

traffic = mean _ service _ time mean _ arrival _ rate _ for _ service

for speech services the capacity of a mobile radio network can be defined as:

capacity = traffic / area

Expanding the above expressions leads to:

traffic ff traffic ff channels carriers sites = area channel carrier site area
50

Capacity p y considerations ( (2) )

Using the definition of cluster size for homogeneous networks as the number of sites where each radio frequency carrier is used exactly once once, the following equation can be derived:
carriers total _ No .of f _ carriers = = site cluster _ size bandwidth carriers bandwidth cluster _ size

By y combining g the equations: q

capacity =

traffic channels carriers 1 sites bandwidth channel carrier bandwidth cluster _ size area

51

Capacity p y considerations ( (3) )

traffic/channel - physical channel (system) load; this factor takes into account the fact that a channel usually cannot be fully loaded loaded. channels/carrier - system dependent parameter (GSM family: 8 TCH for full-rate channels, 16 TCH for half-rate channels, signalling neglected); carriers/bandwidth - system dependent parameter (GSM: 5 carriers per 1 MHz); cluster size - characterizing the frequency reuse in the deployment area, which depends on propagation conditions, required QoS and the network structure; bandwidth - total available frequency bandwidth per operator; sites/area - describes the base station density in the deployment area.

52

Capacity p y considerations ( (4) )

A reasonable definition of the spectral capacity is obtained by relating the capacity to the most severe network investment costs spent for the licensed spectrum and building up the network infrastructure:

spectral _ capacity =

capacity sites bandwidth area

Replacing the capacity leads to:

spectral _ capacity = traffic channels carriers 1 channel carrier bandwidth cluster _ size

53

Capacity considerations (5)

Numerical examples p on spectral p efficiency y
Parameters: Licensed bandwidth: 7.2 MHz; GoS: 1% Scenario 1: Omni cells of cluster 12 Scenario 2: Sector cells of cluster 4x3 (4/12)

4/12 cluster

54

Efficiency y measures
Spectrum efficiency: a measure of how efficiently frequency, time and space are used:

se =

Traffic (Erlang) Bandwidth Area No. of channels/c ell Offered traffic/ch annel = Bandwidth Area

Erlang ) kHz km 2

It depends on:
Number of required channels per cell Cluster size of the interference group

55

Efficiency y( (Utilization related) )

Efficiency =

Traffic nonblocked Capacity Erlangs portions of nonrouted traffic = Number of trunks (channels)

56