Professional Documents
Culture Documents
Introduction To Teletraffic Theory
Introduction To Teletraffic Theory
Introduction To Teletraffic Theory
Examples
Contents
3. Examples 3. Examples
Classical model for telephone traffic (1) Classical model for telephone traffic (2)
• Loss models have traditionally been used to describe (circuit- • Erlang modelled this as a pure loss system (m = 0)
switched) telephone networks – customer = call
– Pioneering work made by Danish mathematician A.K. Erlang (1878-1929) • λ = call arrival rate (calls per time unit)
• Consider a link between two telephone exchanges – service time = (call) holding time
– traffic consists of the ongoing telephone calls on the link • h = 1/µ = average holding time (time units)
– server = channel on the link
• n = nr of channels on the link
µ
1
λ µ
µ
µ
n
3 4
3. Examples 3. Examples
channel-by-channel • The strength of the offered traffic is described by the traffic intensity a
occupation call holding time
6 • By definition, the traffic intensity a is the product of the arrival rate λ
channels
5
4 and the mean holding time h:
3
2
1 a = λh
time – The traffic intensity is a dimensionless quantity. Anyway, the unit of the
call arrival times traffic intensity a is called erlang (erl)
blocked call
nr of channels – By Little’s formula: traffic of one erlang means that one channel is occupied
occupied on average
nr of channels
6
5 • Example:
4
3 – On average, there are 1800 new calls in an hour, and the average holding
2 time is 3 minutes. Then the traffic intensity is
1
0
traffic volume time a = 1800 ∗ 3 / 60 = 90 erlang
5 6
3. Examples 3. Examples
• In a loss system some calls are lost • In a loss system each call is either lost or carried. Thus, there are
– a call is lost if all n channels are occupied when the call arrives three types of call rates:
– the term blocking refers to this event – λoffered = arrival rate of all call attempts
• There are two different types of blocking quantities: – λcarried = arrival rate of carried calls
– Call blocking Bc = probability that an arriving call finds all n channels – λlost = arrival rate of lost calls
occupied = the fraction of calls that are lost
– Time blocking Bt = probability that all n channels are occupied at an λoffered λcarried
arbitrary time = the fraction of time that all n channels are occupied
λlost
• The two blocking quantities are not necessarily equal
– Example: your own mobile
λoffered = λcarried + λlost = λ
– But if calls arrive according to a Poisson process, then Bc = Bt
• Call blocking is a better measure for the quality of service experienced λcarried = λ (1 − Bc )
by the subscribers but, typically, time blocking is easier to calculate
λlost = λBc
7 8
3. Examples 3. Examples
• The three call rates lead to the following three traffic concepts: • System capacity
– Traffic offered aoffered = λofferedh – n = number of channels on the link
λoffered λcarried
– Traffic carried acarried = λcarriedh • Traffic load
– Traffic lost alost = λlosth λlost – a = (offered) traffic intensity
• Quality of service (from the subscribers’ point of view)
aoffered = acarried + alost = a – Bc = call blocking = probability that an arriving call finds all n channels
acarried = a (1 − Bc ) occupied
• Assume an M/G/n/n loss system:
alost = aBc – calls arrive according to a Poisson process (with rate λ)
• Traffic offered and traffic lost are hypothetical quantities, but – call holding times are independently and identically distributed according to
traffic carried is measurable, since (by Little’s formula) it corresponds any distribution with mean h
to the average number of occupied channels on the link
9 10
3. Examples 3. Examples
• Then the quantitive relation between the three factors (system, traffic, • Assume that there are n = 4 channels on a link and the offered traffic is
and quality of service) is given by Erlang’s formula: a = 2.0 erlang. Then the call blocking probability Bc is
an 24 16
4! 24 2
Bc = Erl(n, a ) := n! Bc = Erl( 4 , 2 ) =
22 23 24
=
1+ 2 + 4+8 + 16
=
21
≈ 9 .5 %
n i 1 + 2 + 2! + 3! + 4! 2 6 24
∑ ai!
i =0 • If the link capacity is raised to n = 6 channels, then Bc reduces to
• Given the quality of service requirement that Bc < 1%, the required • Given the capacity n = 20 channels, the required quality of service
capacity n depends on the traffic intensity a as follows: 1 − Bc depends on the traffic intensity a as follows:
80 0.8
60 0.6
capacity n quality of service
40 1 − Bc 0.4
20 0.2
20 40 60 80 100 20 40 60 80 100
13 14
traffic a traffic a
3. Examples 3. Examples
• Given the traffic intensity a = 15.0 erlang, the required quality of service • Model for telephone traffic
1 − Bc depends on the capacity n as follows: • Packet level model for data traffic
• Flow level model for elastic data traffic
1 − Bc (n) = 1 − Erl(n,15.0) • Flow level model for streaming data traffic
1
0.8
0.6
quality of service
1 − Bc 0.4
0.2
10 20 30 40 50
15 16
capacity n
3. Examples 3. Examples
Packet level model for data traffic (1) Packet level model for data traffic (2)
• Queueing models are suitable for describing (packet-switched) data • This can be modelled as a pure queueing system with a single server
traffic at packet level (n = 1) and an infinite buffer (m = ∞)
– Pioneering work made by many people in 60’s and 70’s related to – customer = packet
ARPANET, in particular L. Kleinrock (http://www.lk.cs.ucla.edu/)
• λ = packet arrival rate (packets per time unit)
• Consider a link between two packet routers
• L = average packet length (data units)
– traffic consists of data packets transmitted along the link
– server = link, waiting places = buffer
• C = link speed (data units per time unit)
– service time = packet transmission time
R • 1/µ = L/C = average packet transmission time (time units)
R R
R µ
λ
17 18
3. Examples 3. Examples
19 20
3. Examples 3. Examples
Example Delay
• Consider a link between two packet routers. Assume that, • In a queueing system, some packets have to wait before getting served
– on average, 50,000 new packets arrive in a second, – An arriving packet is buffered, if the link is busy upon the arrival
– the mean packet length is 1500 bytes, and • Delay of a packet consists of
– the link speed is 1 Gbps. – the waiting time, which depends on the state of the system upon the
• Then the traffic load (as well as, the utilization) is arrival, and
– the transmission time, which depends on the length of the packet and the
ρ = 50,000 ∗1500 ∗ 8 / 1,000,000,000 = 0.60 = 60% capacity of the link
• Example:
– packet length = 1500 bytes
– link speed = 1 Gbps
– transmission time = 1500*8/1,000,000,000 = 0.000012 s = 12 µs
21 22
3. Examples 3. Examples
• System capacity • Then the quantitive relation between the three factors (system, traffic,
– C = link speed in kbps and quality of service) is given by the following formula:
• Traffic load
– λ = packet arrival rate in pps (considered here as a variable)
Pz = Wait(C , λ ; L, z ) :=
– L = average packet length in kbits (assumed here to be constant 1 kbit) λL exp(−( C − λ ) z ) = ρ exp(− µ (1 − ρ ) z ), if λL < C ( ρ < 1)
• Quality of service (from the users’ point of view) C L
– Pz = probability that a packet has to wait “too long”, i.e. longer than a given 1, if λL ≥ C ( ρ ≥ 1)
reference value z (assumed here to be constant z = 0.00001 s = 10 µs)
• Note:
• Assume an M/M/1 queueing system:
– The system is stable only in the former case (ρ < 1). Otherwise the number
– packets arrive according to a Poisson process (with rate λ)
of packets in the buffer grows without limits.
– packet lengths are independent and identically distributed according to the
exponential distribution with mean L
23 24
3. Examples 3. Examples
• Assume that packets arrive at rate λ = 600,000 pps = 0.6 packets/µs • Given the quality of service requirement that Pz < 1%, the required link
and the link speed is C = 1.0 Gbps = 1.0 kbit/µs. speed C depends on the arrival rate λ as follows:
• The system is stable since
C (λ ) = min{c > λL | Wait(c, λ ;1,10) < 0.01}
ρ = λCL = 0.6 < 1
2
• The probability Pz that an arriving packet has to wait too long (i.e. 1.75
longer than z = 10 µs) is 1.5
3. Examples 3. Examples
• Given the link speed C = 1.0 Gbps = 1.0 kbit/µs, the quality of service • Given the arrival rate λ = 600,000 pps = 0.6 packets/µs, the quality of
1 − Pz depends on the arrival rate λ as follows: service 1 − Pz depends on the link speed C as follows:
0.8 0.8
0.6 0.6
quality of service quality of service
1 − Pz 0.4 1 − Pz 0.4
0.2 0.2
• Model for telephone traffic • Sharing models are suitable for describing elastic data traffic at flow
• Packet level model for data traffic level
• Flow level model for elastic data traffic – Elasticity refers to the adaptive sending rate of TCP flows
– This kind of models have been proposed, e.g., by J. Roberts and his
• Flow level model for streaming data traffic researchers (http://perso.rd.francetelecom.fr/roberts/)
• Consider a link between two packet routers
– traffic consists of TCP flows loading the link
R
R R
R
29 30
3. Examples 3. Examples
Flow level model for elastic data traffic (2) Traffic process
transfer time
• The simplest model is a single server (n = 1) pure sharing system flow duration with full link rate
with a fixed total service rate of µ
extra delay
– customer = TCP flow = file to be transferred
• λ = flow arrival rate (flows per time unit)
time
• S = average flow size = average file size (data units) flow arrival times
– server = link number of flows in the system
• C = link speed (data units per time unit) 4
3
– service time = file transfer time with full link speed 2
• 1/µ = S/C = average file transfer time with full link speed (time units) 1
0
time
relative transmission rate for single flows
∞ µ
1
λ
1/2
1/41/3
0
31 time 32
3. Examples 3. Examples
• The strength of the offered traffic is described by the traffic load ρ • Consider a link between two packet routers. Assume that,
• By definition, the traffic load ρ is the ratio between the arrival rate λ – on average, 50 new flows arrive in a second,
– average flow size is 1,500,000 bytes, and
and the service rate µ = C/S:
– link speed is 1 Gbps.
λ λS • Then the traffic load (as well as, the utilization) is
ρ= =
µ C
ρ = 50 ∗1,500,000 ∗ 8 / 1,000,000,000 = 0.60 = 60%
– The traffic load is (again) a dimensionless quantity
– It tells the utilization factor of the server
33 34
3. Examples 3. Examples
• In a sharing system the service capacity is shared among all active • System capacity
flows. It follows that all flows get delayed (unless there is only a single – C = link speed in Mbps
active flow) • Traffic load
• By definition, the ratio between the average flow size S and the average – λ = flow arrival rate in flows per second (considered here as a variable)
total delay D of a flow is called throughput θ, – S = average flow size in kbits (assumed here to be constant 1 Mbit)
θ =S/D • Quality of service (from the users’ point of view)
– θ = throughput
• Example: • Assume an M/G/1-PS sharing system:
– S = 1 Mbit
– flows arrive according to a Poisson process (with rate λ)
– D=5s – flow sizes are independent and identically distributed according to any
– θ = S/D = 0.2 Mbps distribution with mean S
35 36
3. Examples 3. Examples
• Then the quantitive relation between the three factors (system, traffic, • Assume that flows arrive at rate λ = 600 flows per second and the link
and quality of service) is given by the following formula: speed is C = 1000 Mbps = 1.0 Gbps.
• The system is stable since
C − λS = C (1 − ρ ), if λS < C ( ρ < 1)
θ = Xput(C , λ ; S ) := ρ = λCS = 1000
600 = 0.6 < 1
0, if λS ≥ C ( ρ ≥ 1)
• Throughput is
• Note:
– The system is stable only in the former case (ρ < 1). Otherwise the number θ = Xput(1000,600;1) = 1000 − 600 = 400 Mbps = 0.4 Gbps
of flows as well as the average delay grows without limits. In other words,
the throughput of a flow goes to zero.
37 38
3. Examples 3. Examples
• Given the quality of service requirement that θ ≥ 400 Mbps, the • Given the link speed C = 1000 Mbps, the quality of service θ depends
required link speed C depends on the arrival rate λ as follows: on the arrival rate λ as follows:
C (λ ) = min{c > λS | Xput(c, λ ;1) ≥ 400} = λS + 400 θ (λ ) = Xput(1000, λ ;1) = 1000 − λS , λ < 1000/S
1400 1000
1200
800
1000
800 600
link speed C throughput θ
(Mbps) 600 (Mbps) 400
400
200
200
200 400 600 800 1000 200 400 600 800 1000
39 40
arrival rate λ (flows per second) arrival rate λ (flows per second)
3. Examples 3. Examples
• Given the arrival rate λ = 600 flows per second, the quality of service θ • Model for telephone traffic
depends on the link speed C as follows: • Packet level model for data traffic
• Flow level model for elastic data traffic
θ (C ) = Xput(C ,600;1) = C − 600 S , C > 600 S • Flow level model for streaming data traffic
400
350
300
250
throughput θ 200
(Mbps) 150
100
50
3. Examples 3. Examples
Flow level model for streaming CBR traffic (1) Flow level model for streaming CBR traffic (2)
• Infinite system is suitable for describing streaming CBR traffic at flow • Model: an infinite system (n = ∞)
level – customer = UDP flow = CBR bit stream
– The transmission rate and flow duration of a streaming flow are insensitive • λ = flow arrival rate (flows per time unit)
to the network state
– service time = flow duration
– This kind of models applied in 90’s to the teletraffic analysis of CBR traffic in
ATM networks • h = 1/µ = average flow duration (time units)
• Consider a link between two packet routers • Bufferless flow level model:
– traffic consists of UDP flows carrying CBR traffic (like VoIP) and loading the – when the total transmission rate of the flows exceeds the link capacity, bits
link are lost (uniformly from all flows)
R µ
1
λ µ
R R •
•
R • ∞
43 44
3. Examples 3. Examples
a = λh
time – This may be called traffic intensity (cf. telephone traffic)
flow arrival times – It follows that
lost traffic
total bit rate
(number of flows) R = ar = λhr
45 46
3. Examples 3. Examples
• Then the quantitive relation between the three factors (system, traffic, • Given the quality of service requirement that ploss < 1%, the required
and the quality of service) is given by the following formula capacity n depends on the traffic intensity a as follows:
∞ i
ploss = LR(n, a) := 1a ∑ (i − n) ai! e − a n(a ) = min{i = 1,2,K | LR(i, a ) < 0.01}
i = n +1 100
• Example: 80
– n = 20
60
– a = 14.36 capacity n
– ploss = 0.01 40
20
20 40 60 80 100
49 50
traffic a
3. Examples 3. Examples
• Given the capacity n = 20, the required quality of service 1 − ploss • Given the traffic intensity a = 15.0 erlang, the required quality of service
depends on the traffic intensity a as follows: 1 − ploss depends on the capacity n as follows:
0.8 0.8
0.6 0.6
quality of service quality of service
1 − ploss 0.4 1 − ploss 0.4
0.2 0.2
20 40 60 80 100 10 20 30 40 50
51 52
traffic a capacity n