1. Traffic Modeling

2. What is Traffic Modeling
Statistical Representation of Traffic generated:
From Users to Network
From Network to Users
Within Network itself
What do we usually Model
Packet Inter-arrival behavior
Packet payload behavior
+ Packet time-correlation behavior
Users
Telecommunication
Network
time
Packet Arrival Behavior over Time

3. Why Do We Need Traffic Modeling?
Network Dimensioning
How many devices + which devices?

4. Why Do We Need Traffic Modeling?
QoS Provisioning
What is expected QoS peformance?

5. Why Do We Need Traffic Modeling?
Connection Admission Control
Which connections should be admitted to network?

6. Why Do We Need Traffic Modeling?
Congestion Control
How to deal with network overload conditions?

7. Traffic at Different Layers in the Network
Characteristics of Traffic at each layer is influenced by
Obviously, the layer above it
For sure, the protocol behavior of the layer itself
And even, the operation of the layer below it
Nevertheless, traffic modeling attempts to model the composite behavior of each layer individually
Usually Layers of Emphasis for Traffic Modeling

8. Traffic at the PHY Layer
Sequence of bits streams
Modeling and Simulation at the PHY layer are usually concerned with BER and Channel Quality
Usually model equal probability of 0 and 1 is assumed

9. Traffic at the MAC+LLC Layer
Starting at the MAC layer, packets appear
The MAC & LLC will impact packet sizes through the maximum lengths they impose
Inter-arrival times are affected by random access behavior along with flow control behavior
The traffic models are still pre-dominantly concerned with point-to-point communication

10. Traffic at the Network Layer
Packet Switching Network
Multiple Levels of Data Aggregation
Multiple Services supported by IP
Routing
Access
Clients

11. Traffic at the Transport Layer
TCP or UDP
Sliding Window Flow Control of TCP
Slow Start of TCP
time to
transmit
file
initiate TCP
connection
RTT
request
received
time

12. Traffic at the Application Layer
Application layer protocols have different characteristics
Client-server or peer-to-peer architectures

13. The Internet Minute

14. The Internet of Things

15. Mobile Data Tremendous Traffic Growth
CAGR: Compound Annual Growth Rate
1 Exa Byte = 1 Billion (109) Giga Byte
[Ref] Cisco, “Cisco Visual Networking Index: Global Mobile Data Traffic Forecast Update, 2013–2018”, White Paper Feb .2014

16. 3GPP Traffic Models Examples: HTTP

17. 3GPP Traffic Models Examples: HTTP

18. 3GPP Traffic Models Examples: FTP

19. 3GPP Traffic Models Examples: NRT Video
NRT = Non Real Time

20. 3GPP Traffic Models Examples: SpeechON
Talk
OFF
Silent
Average ON Period Exponentially Distributed with mean 3 sec.
Average OFF Period Exponentially Distributed with mean 3 sec.
Packet Data Rate and Packet Size during ON Period is Constant

21. Parameters for Traffic Modeling
Parameters needed to model traffic of any type
Packet size
Inter-arrival times
Auto-correlation behavior
Packet size might be easier to model. It is mainly subject to protocol behavior
Inter-arrival times are more challenging
Effect of aggregation
Sources of delay within network
Auto-correlations may have significant impact on network performance

22. Classical Traffic Modeling: Poisson Models
Used originally to model arrivals of calls in a telephone network
The distribution is memoryless. Assumes arrivals are independent
Very simple and easy to use
Can be seen as a counting process. Inter-arrival times follow an exponential distribution
Has a single parameter, arrival rate λ
Mean and variance also equal to λ
Aggregation of Poisson processes remain Poisson
Pr 𝑋=𝑘 = λ 𝑘 𝑒 −λ 𝑘!
𝑓 𝑡 =λ 𝑒 −λ𝑡

23. A New Era of Traffic Modeling
In 1989, Leland and Wilson begin taking high resolution traffic traces at Bellcore
Ethernet traffic from a large research lab
100 m sec time stamps
Mostly IP traffic (a little NFS)
Four data sets over three year period
Over 100m packets in traces
Traces considered representative of normal use

24. Why Not Poisson for Web Traffic?
Poisson distribution does not scale the Bursty Traffic properly.
In fine scale, Bursty Traffic Appears Bursty,
In Coarse scale, Bursty Traffic appears smoothed out and looks like random noise.

25. Enhancing the Poisson Model?
Compound Poisson Process
The model is extended to deliver batches of traffic at once.
The inter-batch arrival times are exponentially distributed, while the batch sizes are geometric.
The model has two parameters:
The mean inter-batch arrival time 1/λ
The batch parameters ρ (between 0 and 1)
Thus, mean packet arrival over time period t is tλ/ρ
The model is still essentially Poisson, which is memoryless

26. Enhancing the Poisson Model?
Markov Modulated Poisson Process
Motivated by the need to generate packet arrivals at different rates
A continuous-time Markov chain varies the arrival rate of a Poisson model
Each state in the Markov chain has an associated arrival rate
To determine these parameters, real traffic traces must be used
The model is designed to fit the real trace based on metrics such as:
mean packet arrival rate,
variance-to-mean ratio of the number of arrivals over a short period, or
long-term variance-to-mean ratio of the number of arrivals
Remains dependent on Memoryless type of of distribution

27. Self-Similar Traffic Models
Scales Bursty traffic well, because it has similar characteristics on any scale.
Gives a more accurate pictures due to measured web traffic behavior
Long Range Dependence in the network traffic

28. Example of Distributions Supporting LRD
The Pareto Distribution
𝑓 𝑡 =𝛽 𝛼 𝛽 𝑡 −𝛽−1
α and β are called shape and location parameters
LRD = Long Range Dependence

29. Observations from Practical Measurements
Characterizing aggregate network traffic is hard
Lots of (diverse) applications
Just a snapshot: traffic mix, protocols, applications, network configuration, technology, and users change with time

30. Observations from Practical Measurements
Packet traffic is bursty
Average utilization may be very low
Peak utilization can be very high
Depends on what interval you use!!
Traffic may be self-similar: bursts exist across a wide range of time scales
Defining burstiness (precisely) is difficult

31. Observations from Practical Measurements
Traffic is non-uniformly distributed amongst the hosts on the network
Example: 10% of the hosts account for 90% of the traffic (or 20-80)
Why? Clients versus servers, geographic reasons, popular ftp sites, web sites, etc.

32. Observations from Practical Measurements
Well over 90% of the byte and packet traffic on most networks is TCP/IP
By far the most prevalent
Often as high as 95-99%
Most studies focus only on TCP/IP for this reason

33. Observations from Practical Measurements
Traffic is bidirectional
Data usually flows both ways
Not JUST acks in the reverse direction
Usually asymmetric bandwidth though
Pretty much what you would expect from the TCP/IP traffic for most applications

34. Observations from Practical Measurements
Packet size distribution is bimodal
Lots of small packets for interactive traffic and acknowledgements
Lots of large packets for bulk data file transfer type applications
Very few in between sizes