4.	Simulation Model

A discrete event world view simulation model was developed using event graphs 
[Schruben, 1983] and implemented using the SLAM II simulation language [Pritsker, 
1987].

The model is comprised of  the following events:

1.	Receipt of a request for a URL at the client.
	If the URL is not in the cache of the PROXY server.
	2.	Start initial contact between the PROXY server and HTTP server 
		(start of  conditional get).
	3.	End contact between the PROXY server and HTTP server
		(end of conditional get).
		If the conditional get indicates that the URL must be retrieved
		4.	Start transmission to the PROXY server from the HTTP 
			server.
		5.	End transmission to the PROXY server from the HTTP 
			server.
		Endif
	Endif
6.	Start transmission to the CLIENT from the PROXY server.
7.	End transmission to the CLIENT from the PROXY server.

A time ordered sequence of requests can be read from one or more CERN log files or 
generated by sampling from a time between arrivals distribution.  Since data 
describing the time for conditional gets is not yet available, the total time between 
events 2 and 5 was described with the regression equation previously discussed.  A 
URL not in the cache of the proxy server is always retrieved.  The times between 
events 1 and 2 and events 6 and 7 are modeled as constants.  The time to retrieve a 
URL in the proxy server cache and transmit it to the client is modeled as a function of 
file size using a regression equation as was previously discussed.

Capabilities for reading CERN log files, TCPDump files, levels of experimental 
factors, and general simulation parameters such as the coefficients of the regression 
equations for transfer times are included in the SLAM II implementation of the 
model.  Values for quantities such as times at which URL requests occurred, file 
transfer times, and file sizes can be read from input files or generated as random 
samples from distributions.  Multiple log files with sequences of URL requests can be 
read and merged with respect to the time of the request.  An experiment definition 
file allows multiple, user specified combinations of the levels of the factors previously 
discussed to be simulated in one run.  An options file includes the list of the names of 
log files, a selection of simulation outputs, the time interval for the simulation, the 
number of replicates for each combination of experiment factor levels, and the 
coefficients of the transfer time regression equations.

The simulation estimates the performance measures defined previously for each 
experiment.  A file of factor levels and their resulting performance measures is 
created.  The format of this file is suitable for input to a spreadsheet program for 
further analysis.  Summary statistics and histograms are always computed for the 
following quantities.



  1.  Time from request to end of file transfer
  2.  File transfer rate
  3.  Cache hit Rate
  4.  Cache hit Rate from input file
  5.  File size from input file
  6.  Transmission rate from input file
  7.  Time in cache - all files
  8.  Time in cache - end of simulation
  9.  File size in cache - end of simulation
10.  Number of bytes in the cache

In the options input file, the user can specify the creation of traces of individual files 
and the following performance measures:

  1.  File transfer time trace
  2.  Hit rate trace
  3.  File transfer rate trace
  4.  End of simulation cache contents trace - time in cache
  5.  File time in cache trace


5.3	Simulation Model Validation and Verification

Verification assures that the model as specified on paper, the event graph model, and 
the model as implemented in the computer program, the SLAM II model are 
equivalent.  The events in the SLAM II correspond one-to-one with the events in the 
event graph.  In addition, verification involves the standard activities associated with 
debugging a computer program.  Validation tasks discussed in the next paragraph 
confirm the verification of the model.

Validation assures that the model and the experiments conducted with the model 
effectively represent reality [Sargent, 1992].
[Marc: paragraph you wrote here]


8.2	Simulation

The SLAM II simulation model (WebSim), along with sample input and results files, 
is available in executable form for Windows.  The SLAM II event model in 
FORTRAN is available for modification but requires the SLAM II or 
SLAMSYSTEM software for execution.

References:

Pritsker, A. A. B.,1987, Introduction to Simulation and SLAM II, 3rd ed., Halsted, 
New York.  

Sargent, R. G., 1992, "Validation and Verification of Simulation Models," in 
Proceedings of the 1992 Winter Simulation Conference, J. J. Swain, D. Goldsman, 
R. C. Crain, and J. R. Wilson, eds., Association for Computing Machinery, New 
York, pp. 104 -114.

Schruben, L., 1983, Simulation Modeling with Event Graphs, CACM, 26, 957-963.