How I tested

Stressing a Web server is simple in theory: Just create some scripts that re-create the operations a typical user would make while accessing the site. But unless you happen to have a few hundred or thousand client computers lying around, simulating enough simultaneous users to create realistic loads on a Web server can be very difficult. RadView Software’s WebLoad solves this problem. The WebLoad software can simulate thousands of simultaneous users accessing a Web site, using a controller system and additional load generators. Rather than try to pile up enough PCs to use as load generators, I also used Ixia’s TXS4 traffic generator. Ixia is better known for its “bit-blaster” network-traffic-generation functionality, but the TXS4 is a module that runs Linux and can create layer 7 traffic, including running RadView’s WebLoad load generator. Using the Ixia hardware, I was able to generate 1,000 simulated clients from a single system.

I created a Web site with lots of graphics and a section secured with SSL. The most basic test simply accessed the home page of the network, producing quick response times and lots of hits per second. I then moved on to more complex scripts that viewed various pages, then logged in through SSL and viewed more pages. These scripts produced much higher loads per simulated user, enabling me to see more difference among the servers while they were running Zeus Web Server. I was unable to generate enough load on the Zeus server with my 1,000-client WebLoad license to make the server unstable or to cause it to fail.

The numbers you’ll see in the tables in the main article show the number of virtual clients; the round-trip time (in seconds), which is the average for all the clients making the request; and the number of hits per second, which is the number of HTTP requests answered in a second, not necessarily the number of pages viewed per second.