In today's economy, e-tailers can't afford to have customers click away to a competitor because of slow or faulty website performance. To prepare for the holiday rush, e-tailers need to address some key application performance issues -- site responsiveness/speed, capacity, and the worst-case scenario, catastrophic failure.
A critical part of the getting-ready process is "making sure IT understands what business is expecting from the site prior to the rush," said Steve Kansa, senior product manager and application performance management (APM) expert at Detroit-based Compuware Corp.
"Once IT understands the requirements, it's a matter of building out the application and environment to support that and verifying that what's built supports what the requirements are," he said.
That includes nailing down issues such as how the site is supposed to function, expected volumes, and expected volumes at peak times.
There's almost no margin for error in this competitive landscape, said Mark Sarbiewski, senior director of product marketing for HP Software's Business Technology Optimization. Performance is especially critical as organizations "make sites more interactive with Web 2.0 and rich clients. There's more logic on the client side, so you can have performance problems there."
Once IT captures what business wants to accomplish, Kansa said, "it's a matter of testing and optimizing as much as possible. With the holiday rush, you have a defined end date. As you build the application, you want to do whatever you can proactively to make sure performance is built in and you start tuning [in development]. As you get closer to launch, you've rooted out the core issues and move to the load testing phase. So you need a tool that tells you not only the performance problem, but the root cause, to allow the infrastructure teams to make changes that you need and still hit the go-live date."
End-user experience counts most
In terms of performance, the end user's experience of the transaction response time can be a deal breaker.
"When you click on something, how fast does that respond to the end user? It's really not how fast messages go back and forth, which is interesting to know, but the most important element is how it appears to the end user," Sarbiewski said. "The end-user experience is the measure of whether customers will be happy."
Organizations also need to think about performance over both the WAN and the Internet, Kansa said.
"You want a tool that enables you to determine how performance is in California vs. New York, or if someone hits it with a dialup link vs. a fast cable modem link," he said. The application "may work great in a test environment, but if you haven't looked at how it performs across the Internet, you're missing an entire dynamic."
Kansa added that organizations also need a testing environment that will support high-volume load testing. "In a lot of cases you may test in the online production system, but it gets trickier as you get closer to the holidays," he said.
Understanding capacity needs ahead of time is critical, Sarbiewski said.
"Preparing for the Friday after Thanksgiving or the Monday following, or announcing a killer sale, how much capacity can you deal with is one of things that becomes a challenge to understand," he said. "And you can't do it manually; that's where you need software to simulate a realistic load and ramp up to the absolute capacity you need. Inevitably something falls over, but if you find that it always starts failing in one part of the overall application first, say the Web server starts to get choked, you can begin to optimize the worst-performing pieces under load. You will save tons of money right from the get-go."
Once a website or an update goes live, continuing to monitor performance is important, particularly for preventing catastrophic failure.
"Maybe you didn't have time to test all the scenarios or get up to the expected user volume and you missed something that allows the application to fail," Kansa said. "Now you need to isolate that ASAP. Ideally you have a solution for monitoring the components and the end-user experience that tell you when you're starting to experience slow performance, so you can drill down in real time to find the problem and root cause."
Sarbiewski added, "Rarely do systems go out like a light. There's usually some churn, so you set up a threshold before people get angry. You need to know when to add capacity. You want to know way before customers are unhappy or leaving the site so you can take action."
Tools to monitor application performance
In terms of choosing APM tools, Jasmine Noel, founder and partner at Ptak, Noel & Associates LLC in New York, said organizations should look to get two sets of views.
"One is to understand exactly what the customer is seeing, so an outside-in view," she said. "Another [is to provide] some sense of how the health of the infrastructure is supporting that. The melding of those two sets of tools together is what the APM manager wants on his desktop."
An APM manager who is proactive will also endear IT to the business, Noel said. Say the APM manager notices patterns of behavior that distinguish an annoyed user who abandons a shopping because of site performance from a returning user who abandons a shopping cart because they're thinking about the purchase decision.
"You know that customer bought before, so maybe if you give them a 5% coupon it will close the deal. If an APM manager goes to a business manager with that kind of information, they're going to fall in love with IT. It's all there buried in the logs of these application and Web servers."