"We had a lot of system problems in 2003-something didn't act right and the server or database would go down," said Shawn Schewgman, vice president of technology for Overstock.com, Salt Lake City, Utah.
That's the stuff of ulcers and extremely sudden job changes at any e-commerce company, much less the Internet's sixth largest retailer, according to Comscore Media Metrix. Overstock.com has doubled in size every year for the past five years and tallied up $300 million in sales in 2003. "It's difficult to manage that kind of growth with more than 1 million unique visitors per day," Schwegman noted.
In diagnosing the previous season's troubles, Overstock.com said its performance problems stemmed from the network attached storage (NAS) it was using. "We beat the hell out of the Network Appliance boxes we were using and they could not keep up," Schwegman explained. "NAS is just not ready for high-end database applications."
Since smooth, lightning-fast interaction between storage and databases is the fuel for any sizable e-commerce site, Schwegman knew he had to make some changes. So in early 2004, he began gathering information from vendors so that there would be no repeat of poor performance in the next end-of-year holiday shopping season.
After discussions with EMC, Hitachi Data Systems, IBM and his incumbent NAS vendor Network Appliance, Overstock.com opted for a 70-terabyte EMC Symmetrix DMX platform, as well as PowerPath software for workload balancing and EMC's TimeFinder/Mirror application for creating online copies of production databases. Schwegman said the company worked closely with EMC for a few months prior to deployment to fine-tune the integration between storage resources and an Oracle 10G database environment, which runs on more than 120 Dell servers using Red Hat Linux.
The company credits the changeout with pages that loaded 77 percent faster, a boon to customers and the retailer alike. Schwegman also said that the company's reporting database gets refreshed daily, and they've reduced the replication time to two hours from the 13 hours it took before, which means faster analytics and reporting data for internal use.
"Performance and reliability were the pain points we wanted to solve," Schwegman noted. "NAS has so much IP overhead, and the SAN is more reliable and just so much faster."