How we did it

Our emphasis during testing was on accelerating typical public-facing Web sites and applications. We used a pair of Dell Pentium 4 servers with 1G byte of RAM running Microsoft's Internet Information Server Version 6 as our primary test devices. The servers were loaded with generic tests cases and dynamic pages written in ColdFusion and Active Server Pages. We used a variety of browsers for testing including Netscape 4.x, 7.x, Mozilla/Firebird, Opera 7 and Safari. We typically used HTTP analysis tools such as HTTPWatch for Internet Explorer or LiveHeaders for Mozilla to carefully analyze responses from the front-end devices.

For our test cases we used a variety of unit tests composed of just HTML; HTML and Cascading Style Sheets; HTML with CSS and JavaScript; and these text formats with a variety of different image sets to simulate realistic page content mixture. To verify the tests we also grabbed the files from home pages of five sites including Amazon, Yahoo, UCLA, BBC and The White House on a June day. We performed the most extensive tests and report on the BBC page because it offered an average mix of content and execution. The others tests provided identical overall results as far as the rankings resulting from the BBC tests. Depending on content mix, compression and caching mileage will certainly vary, sometimes significantly.

Where necessary for testing features we would generate synthetic load using both open source programs such as httperf and WAPT 3.0 for Windows. However, the goal for synthetic load generation was for feature exploration and server saturation and not front-end device saturation because that will be covered in a separate test.

To continue reading, register here and become an Insider. You'll get free access to premium content from CIO, Computerworld, CSO, InfoWorld, and Network World. See more Insider content or sign in.

Our emphasis during testing was on accelerating typical public-facing Web sites and applications. We used a pair of Dell Pentium 4 servers with 1G byte of RAM running Microsoft's Internet Information Server Version 6 as our primary test devices. The servers were loaded with generic tests cases and dynamic pages written in ColdFusion and Active Server Pages. We used a variety of browsers for testing including Netscape 4.x, 7.x, Mozilla/Firebird, Opera 7 and Safari. We typically used HTTP analysis tools such as HTTPWatch for Internet Explorer or LiveHeaders for Mozilla to carefully analyze responses from the front-end devices.

For our test cases we used a variety of unit tests composed of just HTML; HTML and Cascading Style Sheets; HTML with CSS and JavaScript; and these text formats with a variety of different image sets to simulate realistic page content mixture. To verify the tests we also grabbed the files from home pages of five sites including Amazon, Yahoo, UCLA, BBC and The White House on a June day. We performed the most extensive tests and report on the BBC page because it offered an average mix of content and execution. The others tests provided identical overall results as far as the rankings resulting from the BBC tests. Depending on content mix, compression and caching mileage will certainly vary, sometimes significantly.

Where necessary for testing features we would generate synthetic load using both open source programs such as httperf and WAPT 3.0 for Windows. However, the goal for synthetic load generation was for feature exploration and server saturation and not front-end device saturation because that will be covered in a separate test.

Read more about data center in Network World's Data Center section.