[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

http traffic generator



Tom,

        I was pleased to see your implementation
of B. Mah's "Empirical Model of HTTP Network Traffic"
which I found on http://http.cs.berkeley.edu/~tomh/wwwtraffic.html.

	Unfortunately it did not run on my Sun Ultra-30 
with Solaris 2.6. When debugging it seemed as if something was not
correctly initialized. In tcp-sink.h a comment says that the variable
"save_" was placed to stash a saved packet while delaying. The
interrupts observed were caused by a reference to this variable when not
initialized. However, TcpSink::reset which initializes this variable was
called once when starting to run httpTest.tcl. I congesture that this is
not sufficient and added some resets to the httpModel.tcl script. 

***	Since then everything seems to run. 

	Would you please be so kind to verify the few changes that I made to
this script?

Enjoy.

- UF   
	

P.S:
There is a couple of more things about this implementation,
which I would like to ask:
1. The CDF quantization for request/reply size is equal to a
   "there is a largest file in system" assumption. 
   I assume that the size of this largest file somehow 
   relates to the maximum burst rate when such http traffic 
   is aggregated.
   Is that so? - Has this been verified by measurements?
2. Your implementation is based on http 1.0. However, 
   1.1 is common. Does this affect patterns of 
   network traffic? If so, how?

-- 
Ulrich Fiedler   
[email protected]  http://www.tik.ee.ethz.ch/~fiedler 
ETHZ, TIK, ETZ G64.1, CH-8092 Zurich
Phone +41 1 632 7009   Fax   +41 1 632 1035

httpModel.tcl