NEW YORK—Like the floor wax that is also a dessert topping, Amazon.com wants to be more than just a retailer. “Amazon is a retailer, a development platform and an outsourcing solution,” said Tom Killalea, the company’s vice president of infrastructure.
Soon, Amazon will be able to add another title to its collection: pure-Linux enterprise. Killalea announced in his keynote presentation at LinuxWorld on Wednesday that Amazon’s last holdout application—the company’s 14-terabyte-plus data warehouse—will be moved over to Linux servers running Oracle’s Real Application Clusters (RAC) software by the end of the second quarter of this year.
Amazon started its move to Linux in 2000, when it switched its Web servers to the open-source operating system. Over the past four years, Killalea said, the Seattle company has moved more and more of its infrastructure from Sun Unix servers to HP ProLiant servers running Linux. In an October 2001 filing with the Securities and Exchange Commission, the company claimed that Linux had allowed it to cut its technology expenses by $16 million—a reduction of nearly 25 percent.
In his hour-long, PowerPoint-driven presentation, Killalea described at length how the company uses Linux as well as the benefits Amazon has already reaped from the OS when it comes to serving its four classes of clientele: retail customers, Amazon merchants, the company’s own logistical network and developers. Killalea touted the performance of the company’s systems, which over the holiday season handled the shipment of more than a million packages per day, and processed more than 20 million inventory updates daily.
He said Amazon’s Web services software development kit has been downloaded by more than 50,000 individuals. Amazon uses load-balanced Linux Web servers to “horizontally scale” its Web presence. And the savings from the switch to Linux have also allowed Amazon to enhance the fault-tolerance of its infrastructure, Killalea said.
“In migrating application servers, some of which were heavy-duty applications running on single systems, the cost savings from Linux made it possible to have active standby systems,” he told the LinuxWorld audience. Amazon uses a homegrown message queuing architecture and Web services to wire together its collection of internally written applications.
In June 2002, Amazon made many of its Web services interfaces public as a software developer kit; the services provide access to the data within Amazon’s entire product catalog and much of the functionality of the site used by both customers and merchants who sell through Amazon. The services can be accessed via both Simple Object Access Protocol (SOAP) and Representational State Transfer (REST), two XML-based Web services standards; Killalea said that about 80 percent of developers have thus far opted for the REST approach to building sites based on Amazon’s services.
He also described the developer culture within Amazon, where the ideal size for development is a “two-pizza” team—one that can feed all its members with two pizzas. “Amazon is a tech company, dominated by software engineers and run by a computer-science grad,” he said.
Much of Killalea’s keynote amounted to a thinly veiled recruiting pitch for additional developers. He polled the audience to see how many attendees had downloaded Amazon’s SDK, and he showed a few samples of applications built with it, including AmazonBrowser, a Java applet that graphically shows the relationship between books and other products in the Amazon catalog, and Baconizer, a Web site that uses the same principles as the game “Six Degrees of Kevin Bacon” to show how two disparate products are related. “I hope to get more of you working with Amazon,” he concluded, “either by downloading the SDK, or by clicking on Amazon.com/jobs.”
