How we did it

We installed each of our products on a Compaq ProLiant DL380 equipped with twin 1-GHz Pentium-III processors, 512M bytes of RAM and an integrated Smart Array controller with an 18G-byte RAID-5 array. The operating system used was Windows 2000 Advanced Server with Service Pack 2 installed. We also installed Exchange 2000 where appropriate to the product.

We ran a bevy of tests to see how well our content was tracked. To check out attachment management, we used three types of files: jpg, mp3 and pdf. They were sent first using a real file name and extension, and then sent again with bogus names. The files were later archived using WinZip 8.0. Another test was zipping that file a second time, thus creating a recursive archive.

Text analysis was run using simple and complex phrases. First, we used three single-word keywords, and defined them in our dictionaries where the product permitted. For more complex testing, we tried various two-word tests. The first was to detect the phrase "propeller head," either as a literal string or detection of both words. Both words had to be found in that order. Our next test was to find mention of a "stock split." The goal was to test contextual analysis to catch all phrases, such as "our stock split happened," "our stock is going to split" and even "the split will happen with our stock."

The final test was another context analysis exercise, but proved to be the most difficult. We wanted to catch all mail that contained the word "breast," unless it was talking about "breast cancer." So, not only were we looking for two words in a context, but also finding both words resulted in a "pass" condition, and finding only one resulted in a "block" condition.

We graded each product on how well it integrated with antivirus packages. Also, we looked for included mechanisms to limit spam and control relaying.

Footers and disclaimers were created based upon destination address or keyword content. First, we tried to create a rule to add a disclaimer if mail was sent to a specific domain. Our next test used a different disclaimer if a keyword was used in the message body.

Each product was looked at for how well you could group logical units of users (such as departments and managers).

Finally, we looked at what kind of reporting options were available and how easy they were to set up. We tried to send out daily status reports and any charts or graphs. Reports and graphs that told of worst offenders or top rules broken were also on our hit list. Performance metrics was the gravy.