Application Performance: it’s important!

Web application performance can be be measured in a lot of different ways, from requests served to outgoing bandwidth to individual files served. Probably the most important metric for performance isn’t one that can be determined via logging, but rather by asking a site user how the site feels.

• Is it fast?
• Is it slow?
• Is it worth your time?

If we look back at Friendster in 2003 or so, they were probably getting requests in the same order of magnitude as eBay or Amazon. If you chose to look at just the number of requests served, or simultaneous users, and never used the service, Friendster would have like one of the top 25 sites online. However, their infrastructure couldn’t handle their exponential growth, and the user experience was unbearably slow. As a result, a pioneering service lost its foothold to a number of copycats, most notably MySpace, as their users defected to greener, faster pastures.

It is difficult to build a web based service to handle millions of users. This is not so much a technology issue, but a business issue. One of the real limiting factors in web application performance is how much hardware you can allocate. It costs money to have a bunch of servers on the ready for a huge in-flow of users, but there’s no guarantee that those users will come. Can a business afford to have a huge amount of excess computing capacity? On the other hand, can it afford not to? So long as development teams are considering performance at a large scale, and have a plan in place for meeting rapidly growing demand in a short time frame (I’m thinking 2-3 days here) if it arises, the best approach is to pay for the bare minimum that runs your service smoothly. Only actual testing of your service will reveal what that bare minimum is and how your service consumes resources. JMeter from the Apache Jakarta Project is a powerful application that can do just that.

Wow, that’s a lot of lead in. The take home point: well designed web applications save money by minimizing hardware expenses and retaining users.