Amazon.com Inc. has two businesses: The one it runs in November and December during the height of the holiday shopping rush, and the one it runs the rest of the year. Building a cost-effective IT infrastructure that can scale to handle the year-end crush while remaining inexpensive during slower months is one of the obstacles Amazon.com's IT staff grapples with, Vice President of Infrastructure Tom Killalea said Tuesday in a presentation at LinuxWorld.
Amazon.com is a vocal evangelist about the benefits of a Linux-based infrastructure. The online retailer began adopting Linux in 2001, first migrating its Web servers to the operating system, then moving the servers running its custom-built applications for handling systems such as order fulfillment and customer management. Its goal was to develop as inexpensively as possible an IT infrastructure reliable and fast enough to meet customer demands.
The company's computational needs are intense, Killalea said. At the most recent holiday season's peak, Amazon.com shipped 1 million packages in a single day, and processed 20 million inventory updates from third parties. While some of Amazon.com's technology is seen by its 37 million customers, much of the company's IT infrastructure is devoted to handling millions of back-end daily operations, and to serving the 550,000 sellers sharing Amazon.com's platform, Killalea said.
So far, Amazon.com has been pleased with the performance of its Linux infrastructure, and with the cost savings generated, according to Killalea. The company started with what he called the easiest Linux entry point for businesses, server-load balancing technology. From there, it added fault-tolerant clustering systems with failover capabilities -- and there Amazon.com began to see the real advantages of using open-source software. Because Linux-based hardware is so cheap, Amazon.com could afford to set up standbys for systems that historically would have been too expensive to protect with redundant hardware, Killalea said.
Amazon.com is now in the throes of the next stage of its Linux adoption: Moving the servers and warehouse for its Oracle database to Linux boxes. Amazon.com's data warehouse is about 14T bytes, and it needs speedy processing of several gigabytes per second of data. Using Oracle Corp.'s Real Application Clusters on Hewlett-Packard Co. ProLiant servers and HP Modular Smart Array storage systems, Amazon.com is trying to hit all three corners of the "fast, reliable, cheap" technology triangle, Killalea said.
"It's difficult, but from the proof-of-concept we've done it looks like we can do it," he said.
While Amazon.com has smoothed out many of the technology challenges involved in building an always-on, worldwide retail system, others remain daunting, Killalea said, citing personalization as one of the biggest problems bedeviling Amazon.com's IT team.
"It's an area where today we're doing a horrible job, as technologists," he said. "We're doing 1 percent as good a job as we hope to be doing in 50 years."
Amazon.com's latency goals call for information to appear before customers within fractions of a second of being summoned. Meeting that goal, while juggling the myriad factors that go into tailoring content, is a problem that pushes the limits of available technology, Killalea said. Demand forecasting is another area where he anticipates improvement through advancing hardware and software capabilities.
Killealea strayed from the Linux theme toward the end of his speech, when he touted the functionality made possible through Web services and associated standards such as XML (Extensible Markup Language) and SOAP (Simple Object Access Protocol).
Amazon.com released in July 2002 the first version of a software development kit allowing developers to interact with Amazon.com systems exposed through Web services. That kit has now been downloaded more than 50,000 times, Killalea said, and Amazon.com has a list of hundreds of applications created to work with its system -- from tools for sellers to help them better manage Amazon.com inventory, to more whimsical toys such as http://www.baconizer.com/, an application that graphically maps connections between any two items in Amazon.com's extensive catalog.