CS 8803 AIA
[2.1 WS] 1
Strengths:
-The main theme and purpose of the paper is to present 

to the users the problems and issues associated with 

the generation of synthetic Web server traffic in a 

testbed environment. These issues and problems with 

the synthetic web traffic generation serves as a basis 

for the alternative and more robust methodology of 

generating web traffic presented in the paper. 

- In the first half of the paper until section 4.0, 

the problems with synthetic traffic generation is 

presented to the reader. The authors are very clear 

and concise on the problems. One major and obvious 

problem, there is a limit to the number of clients 

that can be put on a LAN due to economic and financial 

constraints.

- It cannot generate large web traffic. The 

inability to generate web traffic is because the 

traditional synthetic model assumes HTTP requests 

generated are independent of the think time of the 

clients. This a fundamental flaw and the authors correct 

this by making it highly dependent on factors such as 

human users sleep and wake patterns. It is this high 

correlation, that allows the new model to generate high 

traffic that is also burstiness. This concept allows

 now the generated rate of requests from clients to

 exceed the capacity of the server which happens in 

realistic and actual working scenarios on the Internet 

as seen in Figure 2 of the paper. 

-This new methodology by the authors now allow to 

evaluate web servers performance under overload 

conditions. This allows researches and developers to

 take into account the poor performance of Unix 

machines under heavy load by simulating actual 

requests from clients. 

Weakness:
- The authors from a very high level understand the 

nature of traffic that needs to be simulated in a 

testbed environment. This is't conclusive enough. 

The authors dont take into account the characteristics 

of client based traces. The authors perhaps might have 

considered taking into account the actual traces of 

clients on a given subset of the Internet. Based on 

this, by performing statistical analysis on the variance 

and the mean distribution, a simulated client request

 generation could have been generated. 

-The importance of statistical analysis is that a more 

thorough generation of client requests based on actual 

world events, popular culture and economic conditions 

a client requests could be generated. 

-The addition of a router in the network is to allow 

the simulation of the WAN delays. The authors could be

 more precise by making the WAN delay a function of 

routers within a certain tolerance limit. This again

 requires more work and statistical analysis but allows

 future researchers to simulate web traffic accurately 

as a mathematical function within a certain acceptable 

tolerance limit.