A former Amazon manager has shared his story about dropping Sun’s hardware, including SPARC-based workstations and servers, in order to save the company. The move away from SPARCstations was also the start of Amazon Web Services (AWS), host of streaming pillars such as Netflix.
Dan Rose said when he joined Amazon in 1999, Sun’s CEO Dave McNealy was on the elevator to see Jeff Bezos. Rose was the chief of retail and later launched the Kindle. Amazon was in trouble, burning $1 billion a year. The biggest expense was its Sun hardware datacenter.
Sun’s workstations were a no-brainer choice for companies in 1999. “In those days, buying Sun was like buying IBM,” Rose writes in a recent Twitter thread. “Nobody ever got fired for it.”
But Amazon spent big bucks to keep its website up. “Sun’s proprietary stack was expensive and sticky,” Rose says. Like a lot of old vendor hardware, pruning it out of the datacenter was difficult.
“Brand new Sun servers started appearing on eBay for 10 cents on the dollar as venture capital-backed start-ups went out of business,” Rose says. “Amazon could have negotiated a better deal with Sun.” Instead, Amazon dropped the proprietary hardware. Rose said the company came within a few quarters of going bankrupt during the migration.
A risky migration
Linux and Hewlett-Packard hosts became the new backbone for Amazon. But the migration away from Solaris was painful and risky.
“All hands were on deck refactoring our code base, replacing servers, preparing for the cutover,” Rose said. “If it worked, infrastructure costs would go down by 80 percent. If it failed, the website would fall over and the company would die.”
Rose goes on to outline how Amazon’s backbone of its Web Services starts with dropping the Sun hardware. “Amazon nearly died in 2000-2003. But without this crisis, it’s unlikely the company would have made the hard decision to shift to a completely new architecture. And without that shift, AWS may never have happened.”
Shifting applications from Solaris to Linux was the highest pass for Amazon to climb while it made its migration. Although Amazon made a shift to Linux, it landed on a platform that’s essential to emulation engines. For example, Linux is the cradle that holds the Stromasys Charon virtualization lineup. Amazon was forced to go the full route, dropping its legacy OS as well as the costly legacy hardware. Capital expenditures dropped, once Amazon shifted away from the Solaris OS hardware.
Even at 10 cents on the dollar, the Sun hardware would carry ancillary costs. Proprietary hardware often has shadowy expenses. Amazon rebuilt with commodity Intel-based hardware. It needed to drop Solaris because there was no way to emulate SPARC in the early 2000s.
Software-application migrations, the bane of any legacy migration, nearly stalled all Amazon IT installation. “Product development ground to a halt during the transition,” Rose says, “and we froze all the new features for over a year. We had a huge backlog, but nothing could ship until we completed the shift to Linux. I remember an all-hands where one of our engineering VPs flashed an image of a snake swallowing a rat.”