Amazon Web Services has never seen an entire data center fail within its massive global infrastructure footprint, an executive in charge of its cloud storage business, said.
“We’ve never seen the loss of a data center,” Mai-Lan Tomsen Bukovec, VP and general manager of Amazon S3, the storage companion to the AWS cloud compute service EC2, said during a keynote Thursday at the company’s re:Invent conference in Las Vegas.
The world’s biggest cloud provider has had many service outages over the years. Because it’s the world’s biggest cloud, its outages often affect thousands of businesses, including some of the internet’s most visited websites. At the bottom of this article is a long list of AWS outages we’ve covered here on DCK over the years.
But, according to Bukovec, those outages never involved the loss of an entire data center.
If that’s true (and some of the outages listed below cast a shadow of doubt on the claim), such a track record of data center uptime would be rare (even considering that an entire data center doesn’t have to go down to disrupt thousands of customers). Even facilities with the most infrastructure redundancy and the most sophisticated automatic failover systems go down time to time, often due to human error, but sometimes also because of unforeseen failures of the failover systems themselves.
Bukovec said this while describing the lengths AWS infrastructure engineers go to ensure “durability” of customer data. Even though a whole AWS data center has never gone down, the system is designed so that customers can quickly retrieve data stored in any building in case such an outage does occur.
“We worry about it, and we include the scaling and the capacity modeling, and the software systems to handle it, so you don’t have to,” she said.
Other hyperscale cloud platforms also design their infrastructure to continue running when a single data center fails. Facebook intentionally shuts down a data center from time to time to test the platform's resiliency.
Amazon’s biggest cloud rival, Microsoft Azure, cannot make such a claim. Earlier this year, a lightning strike caused an electrical load spike in the utility infrastructure powering an Azure data center in San Antonio, Texas, overwhelming its protection mechanism and shutting the entire facility down.
Every AWS availability region around the world consists of multiple availability zones. “In every zone, you have at least one data center,” Bukovec said. It’s never a room within a building that has other data centers; “it’s a whole building,” she said.
Many availability zones consist of multiple data centers, and while AWS shares publicly the number of zones in each region, it does not disclose the number of facilities in each zone. The cloud provider currently has 57 availability zones across 19 geographic regions. It has announced plans to build 15 more zones and five additional regions.
Here’s some of our coverage of AWS cloud outages over the years: