WWW Proxy Traffic Modeling

1
WWW Proxy Traffic Modeling

• Ghaleb Abdulla
• Dow Chemical Company

2
Objectives

• A comprehensive study of WWW proxy traffic
  characteristics to support caching studies
• Develop an empirical model of request arrival
  rate and bytes rate to a proxy
• Develop usage scenarios to study Web scalability
  using caching, prefetching, and delta encoding

3
Data Collection

• Collect data over the network using tcpdump and
  tools to filter http accesses to remote servers
• Collect data from high and middle schools,
  universities, home users, and commercial sources

4
Workload Characterization (Challenges)

• Collection Issues
• Reliable (check hardware and software regularly,
  develop automated monitors and main repository
  with automated mechanism to collect data from
  distributed resources)
• Longitudinal (to study rate of change, data has
  to be collected over a long period of time)
• Sampling (location, duration)

5
Workload Characterization (Challenges)

• Privacy issues
• develop tools to preserve clients anonymity
• Data analysis and visualization
• no off-the-shelf tools to help in the process
• huge data sets hard to manipulate and visualize

6
Proxy Traffic Characterization

• Most comprehensive characterization study of
  proxy traffic
• 10 different workloads
• collected from different domains
• Characterize parameters identified from the log
  file such as file sizes, types, etc.
• Longitudinal study for the longest workload
  collected

7
Modeling proxy traffic

• We are looking at long term behavior of proxy
  users for caching studies
• Arrival rate and bytes rate display a cyclic
  behavior that can be modeled to support
  prefetching and push of information
• Clients behavior has a deterministic component
  that can be used to predict future usage

8
ACF (Request rate)
9
FFT
10
ACF (Random Signal)
11

Modeling proxy traffic

• Extract the mean
• Extract the deterministic part using Fourier
  analysis
• Model the residual using statistical methods
• Create the full model

12
Conclusions

• Proxy traffic characteristics
• Longitudinal study of the CS workload
• Web traffic is highly co-related
• file sizes and types
• inter-arrival time and bytes rate have a
  deterministic component that can be attributed to
  hourly, daily, and weekly schedules and can be
  modeled using Fourier analysis