Here to serve The Lab checks out mid-level servers for organizations that need a quick Web site. (By Greg Crowe)

BACK WHEN THE World Wide Web was just getting started, creating Web space on one of your own servers was close to impossible. Unless you were an expert, the amount of time and effort it took to make sure all of the correct software was loaded and all of the settings were right would have driven even the most patient network administrator up a wall.

With the introduction of Internet Information Services (IIS), it became increasingly easy to give a Windows-based server Web functionality. With today’s versions — IIS 6.0 on Windows Server 2003 and IIS 7.0 on Windows Server 2008 — adding Web services is as simple as flipping a light switch.

There are times when a network administrator is asked to set up a quick Web site for one purpose or another. The best solution often would be to purchase a server to be dedicated to that project. A few connections, settings and file transfers later, a new Web site is born. We looked at servers from Dell, Fujitsu, Hewlett-Packard, and Supermicro with an eye toward how they would perform as Web servers.

Our performance tests were three-pronged. First, we ran each server through the rigors of our PassMark Professional benchmark program. We managed to produce scores that were often two to three times what a typical desktop PC could do. This was, of course, because of the more powerful processors and greater amounts of memory in servers than in regular PCs. But the 64-bit operating systems found in these servers also make them significantly faster than their 32-bit counterparts, a must for high-end Web serving.

Because updating a Web site usually involves a great deal of file uploads, we also measured transfer times across a network connection both to and from the server. And of course, because of our primary concern was how they performed as Web servers, we timed page loads from computers both inside the network and out on the Internet.

This real-world testing revealed some interesting results, but to get meaningful differences in the page load times for each product, we created a Web page that was as slow-loading as we could make it. In the process, we pulled out every bad Web design practice that we have all learned not to do over the years.

This ugly thing had atrocious graphics tiling in the backgrounds, and some huge image files. The granddaddy of them all was a high-resolution GIF file that was more than 3M in size. In an environment where graphics more than 50K are a big no-no, this monstrous page definitely gummed up the works. All of these factors slowed down the page load times enough to measure the differences with a stopwatch from server to server, even when browsing to it from within the test network.

We also measured upgradability, the capacity the system has for upgrades without having to replace any components. Our ease of access grade assessed how easy it was to get to the various internal locations of the server to perform upgrades or maintenance. We then looked at all the extras the server might offer, such as ports, drive bays and cooling systems, and used that information for the features grade. Lastly, we examined the price compared to how it did in the other categories and arrived at the value grade.