• E-mail this page
• |  Print this page
• | 
• | 
• | 

Cloud Computing Serves Up Occasional Thunderstorms

Last weekend's Amazon S3 outage shows that cloud computing isn't all upside. In theory, outsourcing services to companies that focus on them should make things more scalable and reliable. In reality, it doesn't always seem to work that way.

In a particularly unfortunate coincidence, on Friday I had just begun to set up my company to use S3 for serving out some of our files. When I went to apply some finishing touches on Sunday, the Bucket Explorer tool I was using incorrectly said I wasn't connected to the Internet. Finally I thought to check the Amazon AWS status page and sure enough, Amazon was reporting problems. The outage lasted at least eight hours, from about 9 a.m. to 5 p.m. Pacific time.

In fairness to Amazon, I have seen problems with other Web-based services. We use Google Apps for Domains and have had a couple of hiccups with the IMAP mail service over the course of six months, but they haven't lasted for more than half an hour and we didn't lose any data. But people immediately noticed that their mail wasn't arriving, because Outlook was reporting an error accessing Google's mail server. Similarly, I've tried going to Windows Live Search a couple of times in the past month, only to be greeted with strange errors or slow-loading pages.

The Amazon outage has made me think twice about putting business-critical files on S3, but it hasn't soured me on the service completely. It's a hassle to configure more disk space on a server because you pay for it in big quantum chunks. Amazon's price for S3 is so much better than what we can get with our hosting company that it's worth a risk. However, we won't be doing what sites like Reddit did and host all the site's images on S3. Instead, we'll focus on large downloads. Amazon's ability to function as a BitTorrent tracker is a big plus in those situations.

As more companies start to depend on services "in the cloud" for day-to-day operations and integrate them into critical site functions, I suspect we'll be seeing more of these horror stories. For this incident, Amazon is promising to give a full report of what happened, but it hasn't said when that report will be released. If things work like they should, though, an all-day outage like this should spur Amazon to re-evaluate its procedures and make sure this won't happen again -- certainly not on this scale.

« OSCON, Pt. 2.1: A Few Words With Mark Shuttleworth | Main | VMware 2Q Numbers »