Test gear not just for the lab

You probably outsource Web site testing to a lab, but there's no reason you can't do it yourself. Testing equipment might be a bit expensive, but most likely it will pay for itself with improved customer experience.

To properly test equipment in my lab, I have to make sure I can hit it with enough traffic to simulate a popular Web site on a busy day. The idea is to make sure that a typical user will get a fast enough response, and that the equipment--everything from servers to switches--can handle enough user requests to be economical.

Plenty of labs run by publications, service providers, and manufacturers perform similar tests. But test equipment shouldn't just live in labs. Now that the life or death struggle of your business depends on the performance of your Web site and your customers' interaction with your intranet, you really need to know what those customers experience.

Of course, you can log on yourself and use the site, but that doesn't necessarily tell you a lot. Unless you know how your complete Web site or commerce site and all of its related infrastructure is performing under load, you really can't tell much, except that maybe it works fine without much traffic. But your customers won't always access your site when traffic is low.

So you're faced with a couple of bad alternatives. The first is that you can estimate your needs and hope for the best, but that may mean some customers being trapped in a slow, perhaps non-responsive, Web environment. Another choice is to make everything as big and fast as possible, which yields a great customer experience, but at a big cost.

But there is another choice. You can build your environment and then run a simulated load, matching the traffic patterns and levels you would expect in real life, and then see what happens. One excellent way to accomplish this is to use the same kind of test equipment that I use in the lab, whether it's from the big three--Spirent Communications, Hewlett-Packard, and Ixia Communications--or someone else. The equipment costs anywhere from $5,000 to $150,000.

Admittedly, some of this test equipment can be expensive. But once you've decided to make sure your customers are getting what they expect, you can find other things to test, including your WAN fiber, or even your service level agreements. If your development staff comes up with a new version of your Web site, you can test it to make sure it still performs well under load.

"It really saves money," explains Spirent VP Mark Fishburn, explaining that overbuilding a network can be extremely wasteful. Fishburn also suggested that for many enterprises, the ability to recreate a problem with realistic traffic levels was frequently vital to resolving customer support issues.

The bottom line, however, is that in many cases, your Web site is so important to your company's success that you can't just hope it's performing like it should. You have to be sure, and you have to be sure it stays that way. The only way to do that is to test what you have, test again when you change anything, and test routinely just in case something unknown changes. You can use a Web performance tool such as Network Physics NP-1000 to monitor your traffic, and you should. Don't just test in an isolated environment. Test the actual traffic going to your site. Then you can be confident that you're providing your customers with the best environment you can, and that a poor experience won't drive them to your competitors.

What kind of testing does your company do to ensure your Web site performs well under the duress of high traffic? TalkBack below or e-mail us with your thoughts.