How Dollar Thrifty geared up for summer

Web performance load testing from Gomez helped spot potential site bottlenecks

As the busy summer season was approaching, Dollar Thrifty Automotive Group didn’t want to take any chances on its Web properties. So the car rental company put its two redesigned Web sites through a host of performance tests, including on-demand load tests from Gomez.

As the busy summer season was approaching, Dollar Thrifty Automotive Group didn’t want to take any chances on its Web properties. So the car rental company put its two redesigned Web sites through a host of performance tests, including on-demand load tests from Gomez, before its peak sales season got into full swing.

Specifically, Dollar Thrifty wanted to test how its multi-tiered www.dollar.com and www.thrifty.com sites performed under load -- from its customers’ perspective.

“We wanted to make sure that our production environment could sustain the high traffic volumes that we come across every summer. Not only that, we want to remain proactive in our marketing efforts. We may do an e-mail blast occasionally, for example, that creates additional load on top of already peak situations,” explains Jim Arrowood, director of Web development and architecture at Dollar Thrifty Automotive Group. “Those were the types of loads we wanted to generate from Gomez and then test where we were from a production perspective.”

Gomez’s services combine load generated from the cloud with load generated from the Gomez Last Mile network of 80,000 desktops. This combination let Dollar Thrifty create production-equivalent loads and preview how its Web properties would scale under peak conditions across its key markets.

Internal testing environments typically don’t reproduce capacity on a one-to-one scale but instead rely on traffic ratios that allow IT to judge how systems might perform in production, Arrowood points out. “That’s really the big difference: Gomez is able to do that on the production environment in a safe and controlled way, so you know when it’s going to happen and what it’s going to be doing,” he says.

In addition, the service allowed Dollar Thrifty to expose all elements of its Web application environment, including legacy systems, to load testing.

“It’s a multi-tiered application, so we’ve got front-end servers, middle-tier servers and other, more discrete Web services underneath that, such as systems that provide reservations and rates,” Arrowood says. “All those connections and all those hops have the potential for their own bottlenecks and their own issues. We’re able to pinpoint, with the kind of load we’re able to throw from Gomez, where those things are going to tip over in the production world.”

Dollar Thrifty took pains to model how real users would be using its applications for activities including rate shopping, booking a rental reservation, and modifying or canceling a reservation. “We didn’t just see how many HTTP requests we could throw at a Web server,” Arrowood says. “We built individual tests to model all those things.”

So what was it like actually launching the load tests for the www.dollar.com and www.thrifty.com sites? Arrowood and his colleagues staggered the tests over two nights, beginning at midnight and ending at about 3:00 a.m. each night.

“We basically had a pretty large SWAT team engaged on the bridge when this happened. So we had about 30 engineers on the call, and we had a war room at our headquarters where we had the lead team that was actually pushing the buttons to launch the tests,” he recalls. “Each engineer for each appropriate tier of the application had all of their monitoring and diagnostic tools up and running and they were collecting as much data as they could. Once we completed a test, we took the data and came up with an analysis of where the bottlenecks and pain points were, based on the transaction we were testing.”

In the end, the load tests spotlighted a few performance bottlenecks that Dollar Thrifty remedied prior to its peak sales season. For instance, the tests revealed an imbalance in capacity between the two sites that the company remedied by reallocating servers and adjusting its load balancing. “Based on the results, we’ve re-tuned the environment, we’ve made some capacity changes and modified the configuration of load balancers,” Arrowood says.

Going forward, Arrowood sees the Gomez load-testing services as a complement to Dollar Thrifty’s existing Web development lifecycle testing tools. The services are useful not only when there’s a major Web site redesign but also as a way to confirm that ongoing Web development efforts or network changes, for instance, perform as expected once put into production.

“When you’ve made enough changes to a production environment, it’s always good to take a step back and make sure everything is performing the way you think it is,” Arrowood says.

Join the Network World communities on Facebook and LinkedIn to comment on topics that are top of mind.
Now read: Getting grounded in IoT