1. Network Traffic Modeling
Arthur L. Blais
University of Colorado,
Colorado Springs

2. Introduction Nature of Internet Traffic
Trace-based vs. Analytic Models
Network Traffic Characteristics
Model Data and Distributions
February 18, 2019
Arthur L. Blais - UCCS

3. Nature of Internet Traffic
Highly variable demands.
Request inter-arrival rates.
File sizes and references.
Self-similarity.
February 18, 2019
Arthur L. Blais - UCCS

4. Trace-based Model Uses actual data traces to generate workloads
Advantages
Easy to implement
Disadvantages
Difficult to change or vary
Difficult to find causes of system behavior
February 18, 2019
Arthur L. Blais - UCCS

5. Analytic Model Mathematical models are used to generate workloads
Advantages
Capable of generating different workloads by varying workload characteristics.
Disadvantages
Difficult to construct
Need to identify important workload characteristics to model
Characteristics must be empirically measured
February 18, 2019
Arthur L. Blais - UCCS

6. Network Traffic Characteristics
Client Characteristics
Server Characteristics
Network Characteristics
February 18, 2019
Arthur L. Blais - UCCS

7. Client Characteristics
Request Rates
Sleep Time
Active Time
Inactive OFF Time (think time)
Active OFF Time (embedded references)
Request Sizes
February 18, 2019
Arthur L. Blais - UCCS

8. Server Characteristics
File Sizes
Cache Size
Temporal Locality
Number of Connections
CPU speed
February 18, 2019
Arthur L. Blais - UCCS

9. Network Characteristics
Bandwidth
Routing Path
Hop Count
Buffer Sizes
Packet Loss Rates
February 18, 2019
Arthur L. Blais - UCCS

10. Workload Models Client Models Server Models February 18, 2019
Arthur L. Blais - UCCS

11. Client Workload Model Sleep Time Inactive Off Time
Each Client has a time of day that it is awake
Fixed time assigned to each client so that the request rates from all the clients approximates the total request rate distribution for each hour of the day.
Inactive Off Time
Client think time, the amount of time after a request is received and the time the next request is made.
Pareto Distribution
February 18, 2019
Arthur L. Blais - UCCS

12. Client Workload Model Active Off Time Embedded References
Inter-arrival time for each embedded request
Weibull Distribution
Embedded References
The number of references included with the requested document
Pareto Distribution
February 18, 2019
Arthur L. Blais - UCCS

13. Server Workload Model CPU speed Number of Connections
File Size Distribution
Distribution Body ( <= 9020 bytes )
Lognormal Distribution
Distribution Tail ( > 9020 bytes )
Pareto Distribution
February 18, 2019
Arthur L. Blais - UCCS

14. Client Sleep Time
Approximates the percentage of the Total Request Rates for each hour of the day.
Each client has a wakeup time and a sleep time attribute.
February 18, 2019
Arthur L. Blais - UCCS

15. Client Hourly Request Rate
February 18, 2019
Arthur L. Blais - UCCS

16. Inactive Off Time Time between requests Uses a Pareto Distribution
alpha: a = 1.5
Lower bound: (k) = 1.0
To create a random variant x:
u ~ U(0,1)
x = k / (1.0-u)^1.0/a
February 18, 2019
Arthur L. Blais - UCCS

17. Inactive Off Time
February 18, 2019
Arthur L. Blais - UCCS

18. Active Off Time Time between embedded references
Uses a Weibull Distribution
alpha: a = 1.46 (scale parameter)
beta: b = (shape parameter)
To create a random variant x:
u ~ U(0,1)
x = a ( -ln( 1.0 – u ) ^ 1.0/b
February 18, 2019
Arthur L. Blais - UCCS

19. Active Off Time
February 18, 2019
Arthur L. Blais - UCCS

20. Embedded References Number of references in the requested document
Uses a Pareto Distribution
alpha: a = 1.5
Lower bound: (k) = 1.0
To create a random variant x:
u ~ U(0,1)
x = k / (1.0-u)^1.0/a
February 18, 2019
Arthur L. Blais - UCCS

21. Embedded References
February 18, 2019
Arthur L. Blais - UCCS

22. File Size Distribution
Two Distributions
Body
Lognormal Distribution
Create a lookup table
Tail
Pareto Distribution
alpha: a = 1.5
Lower bound: (k) = 1.0
To create a random variant x:
u ~ U(0,1)
x = k / (1.0-u)^1.0/a
February 18, 2019
Arthur L. Blais - UCCS

23. File Size Distribution - Body
February 18, 2019
Arthur L. Blais - UCCS

24. File Size Distribution - Tail
February 18, 2019
Arthur L. Blais - UCCS

25. Self-similar Traffic High Variability
Request Rate Inter-arrival Time
File Sizes
Negative impact on network performance
Modeling Characteristics
Heavy Tailed Distributions
Significant variability over a wide range of scales
February 18, 2019
Arthur L. Blais - UCCS

26. Self-similar Traffic
February 18, 2019
Arthur L. Blais - UCCS

27. References
Paul Barford and Mark Crovella, Generating Representative Web Workloads for Network and Server Performance Evaluation, Boston University, Technical Paper BU-CS , December 31, 1997
Mark E. Crovella and Azar Bestavros, Self-Similarity in World Wide Web Traffic: Evidence and Possible Causes, Boston University, Technical Paper BU-TR , 1995,1996, 1997
Martin F. Arlitt and Carey L. Williamson, Internet Web Servers: Workload Characterization and Performance Implications, IEEE Transactions on Networking, Vol. 5, No. 5, October 1997
Vern Paxon and Sally Floyd, Wide-Area Traffic: The Failure of Poisson Modeling, Lawrence Berkeley Laboratory and EECS Division, University of California, Berkeley, July 18,1995
February 18, 2019
Arthur L. Blais - UCCS