The stress test requires three machines:
1)  A machine on which the idealized web server runs.  This machine must be 
UnixWare7 since the server binary uses UW7 threads.
2)  A machine on which the proxy server runs.
3)  A machine on which the simulated client requests originate.

Machine 3 generates requests from the proxy server on machine 2 very rapidly.
Machine 2 then proxies these requests, making requests of machine 1 or serving
documents out of its cache depending upon its caching rules.

In order to run theses tests, place the binary named "server" on machine 1
and type "./server -p 80 -n 5", which creates a webserver listening on port
80, having an initial threadcount of 5.  Next, install squid on machine 2.
It will be necessary to turn off access restrictions in order to test squid.
To do this, edit /usr/internet/squid/etc/squid.conf and comment out the line:

http_access deny !localhost

so that it looks like:

#http_access deny !localhost

then start squid with "squid start"

Finally, use tcp-banger2 and urlGen to simulate client requests from machine
3.  In order to generate these requests, try a command line like:
./urlGen 10 1000000 galois.pdev.sco.com |./tcp-banger2 -h xxx.xxx.xxx.xxx -p 8080 -n 16
"galois.pdev.sco.com" should be replaced by the name of machine 1 (the machine on which
"server" is running).
"xxx.xxx.xxx.xxx" should be replaced by the numeric IP address of machine 2 (the machine
running squid)

Here's an explanation of what this all means:
urlGen takes as arguments the mean document size to request (in K), the number of urls to
generate, and the address of the server.  So, for example, 
./urlGen 10 1000000 galois.pdev.sco.com generates one million randomized urls (on stdout) 
that look like:
http://galois.pdev.sco.com/file9k.html
http://galois.pdev.sco.com/file1k.html
http://galois.pdev.sco.com/file10k.html
http://galois.pdev.sco.com/file8k.html
http://galois.pdev.sco.com/file11k.html
http://galois.pdev.sco.com/file1k.html
http://galois.pdev.sco.com/file41k.html
http://galois.pdev.sco.com/file5k.html
http://galois.pdev.sco.com/file18k.html
http://galois.pdev.sco.com/file9k.html

"server" understands these requests and returns the appropriate sized file, served straight
out of memory.  Note that the mean size fed to urlGen is a mean value over an exponential
distribution.

This stream of URLs is fed into tcp-banger, which requests these documents from the proxy
server running on machine 2 (-h xxx.xxx.xxx.xxx) on port 8080 (-p 8080), maintaining 16 
(-n 16) outstanding connections at all moments.

I verify that all requests are served and that there are no errors logged in 
/usr/internet/squid/logs/*.  When first started, you should see a frenzy of output from
server on machine 1.  It writes to stdout every time it gets a request.  The frequency
with which it writes should decrease fairly quickly as squid caches more and more of the
data.
