Traffic Intensity
Traffic Intensity
Virtamo
Traffic intensity
a = T
where = number of carried connections per time unit (arrival rate, call rate) T = mean duration of a connection or holding time Traffic intensity is a bare number, but in order to emphasize the context, one often writes as its unit erlang (E, erl) A.K. Erlang (1878-1929) was the pioneer of traffic theory, which he applied to study telephone systems Traffic intensity describes the mean number of simultaneous call in progress Instead of a "connection" we may consider reservation of any resource (trunk, modem, capacity unit)
J. Virtamo
Example
In a local switch the number of calls in an hour is 1800 The mean holding time of a call is 3 min
a = 1800 x 3 / 60 = 90 erlang
Typical traffic intensities per a single source are (fraction of time they are being used) private subscriber business subscriber mobile phone PBX coin operated phone 0.01 - 0.04 erlang 0.03 - 0.06 erlang 0.03 erlang 0.1 - 0.6 erlang 0.07 erlang
J. Virtamo
blocked trafic
Offered traffic ao
traffic, which would be carried were there no constraints in the system a theoretical concept
Carried traffic ac
traffic that is actually being carried
J. Virtamo
Traffic volume
The amount of traffic carried during a given period of time is called the volume of the traffic
Traffic volume in a period divided by the length of the period is the average traffic intensity in that period
J. Virtamo
Traffic variations
Traffic fluctuates over several time scales trend ( > year)
the overall traffic growth: number of users, changes in the usage traffic predictions give the basis for network planning
The last component is purely stochastic The other variations by large follow a given profile, around which the traffic randomly fluctuates (each day, week, month... is different)
J. Virtamo
J. Virtamo
50
40
Number of customers
30
20
10
10 Time/hour
15
20
25
Traffic of the student modem pool at HUT; working days in October 19972.
1
J. Virtamo
In either case, the intensity of the Poisson process varies as a function of time: (t) The actual arrivals occur at random instants of time according to the Poisson process
the probability for an arrival in the interval (t,t+dt) is (t) dt
J. Virtamo
Note. At no time scale does the word "regular" imply fully deterministic behaviour. (t) fluctuates around the average profile. In addition, the arrival process may exhibit variations induced by some external event
predictable / unpredictable regular / irregular
J. Virtamo
10
Busy hour
It is not practical to dimension a network for the largest traffic peak that may ever occur. For pragmatic dimensioning work, one has developed a computational quantity which tries to adequately describe the peak load, but where singular peaks have been averaged out.
The period of duration of one hour where the volume of the traffic is the greatest Due to several random factors, the traffic fluctuates around its average In order to determine an appropriate dimensioning load, the recommendations define how the busy hour traffic shall be measured In fact, there are several definitions (ITU E.600)
an operator may choose the most appropriate one
J. Virtamo
11
the resolution of the start time of the busy hour may be either a full hour (ADPH-F) or a quarter of an
hour (ADPH-Q)
TCBH (Time Consistent Busy Hour) a period of one hour, the same for each day, which gives the greatest average traffic over e.g. 10
days
FDMH (Fixed Daily Measurement Hour) a predetermined, fixed measurement hour (e.g. 9.30-10.30); the measured traffic is averaged over
e.g. 10 days
aFDMH
aTCBH
aADPH
J. Virtamo
12
Busy hour
The busy hour definitions are further divided according to the used time resolution. For instance,
ADPH-F ADPH-Q resolution of an hour resolution of an quarter of an hour
aADPH-F
aADPH-Q
J. Virtamo
13
Busy hour
fixed ADPH, Average Daily Peak Hour TCBH, Time Consistent Busy Hour FDMH, Fixed Daily Measurement hour
J. Virtamo
14
Quality of service
A network cannot be dimensioned for the worst case peaks. Then, occasionally the requested service is not available or the quality of the service is reduced. The dimensioning has to made according to the stated (statistical) criteria for the quality of service
grade of service (GoS): quality at the call level (e.g. telephone network) quality of service (QoS): quality during a connection or session (e.g. ATM network)
J. Virtamo
15
Waiting system
if connecting the call is not immediately possible, the call may be put in a waiting state a small waiting time does not matter, a user may not notice it at all long waiting times are unacceptable for the users one sets an upper limit to the waiting time, after which the call is blocked the behaviour of a waiting system is described by so called Erlang's C formula
J. Virtamo
16
It is not reasonable to dimension the network for a very small blocking probability, since the call may be unsuccessful due to other reasons with a much higher probability:
B subscriber does not answer B subscriber is busy one has dialled a wrong number
J. Virtamo
17
In other networks than the traditional POTS, the quality of service is described by many other quantities, in place of or in addition to the blocking probability In ATM networks and in packet networks, e.g. the Internet, the following may be important
packet / cell delays delay variation (jitter) the proportion of lost packets / cells the proportion of erroneous packets / cells throughput
J. Virtamo
18
Erlang's formula
Assumptions A loss system: a blocked call is cleared (no reattempts) There are n trunks; any free trunk can be used The arrivals constitute a Poisson process
the arrivals occur at average rate otherwise, the arrivals are completely random this is good model when the calls originate from a large population of independent users
Relates the system (n), the traffic (A) and quality of service (E)
J. Virtamo
19
Example In a modem pool there are n = 4 modems and the offered traffic intensity is A = 2 erlang. What is the probability that a call attempt fails?
Consult precomputed graphs / tables or compute directly from the formula
J. Virtamo
20
When A is small, the number of trunks is manyfold in comparison with the mean load
the load factor (utilization) is small
J. Virtamo
21
Call blocking
the fraction of all calls that are blocked
Fortunately, with the assumptions of Erlang's formula (Poisson arrivals), time blocking and call blocking are the same
J. Virtamo
22
J. Virtamo
23
J. Virtamo
24