Modeling web traffic distribution I have to test some functionalities of a software. I need to simulate the web traffic to a server in terms of the numbers of access. What it is the most suitable function distribution for this task? 
 A: It depends on your workload characterization as well as the system.
Take a loot at Long-lasting transient conditions in simulations with heavy-tailed workloads by Crovella and Lipsky publish at Winter Simulation Conference 1997. Specifically, read the paper, use its references, and also check who reference this paper.
A quick and dirty response would be modeling the interarrival request times as an Exponential distribution. But you should consider your system as well, which implies increase or decrease the number of workload generators (clients) in order to make them produce significant requests capable of impact on the system. You also can check Bench4Q: A QoS-Oriented E-Commerce Benchmark as a reference of how architect your testbed.
As more details of your problem you point, more accurate responses will appear. (my opinion)
A: Wiki Site Traffic_generation_model is a starter. It describes two models, 


*

*the more complicated 3GPP2 model, and

*the easier 802.16 model
The latter is a combination of three processes:

  
*
  
*Interrupted Poisson Process (IPP)
  
*Interrupted Discreet Process (IDP)
  
*Interrupted Renewal Process (IRP)
  

They are run in parallel to generate traffic. The main point for you is the IPP, which is used to simulate HTTP and FTP traffic. Short summary:
4 processes, each with different parameters. Each process has two states


*

*ON state, where it sends traffic

*OFF state, where it is idle


There is a switch probability given by a poisson distribution. You can model this by getting the time to the next switch from an exponential distribution.
In the ON state, messages are sent (one-directionally)
