AWS Regions: Not All Created Equal

During a recent conversation with some mates, the topic of AWS stability came up. I’ve had nigh on zero complaints with AWS (except with their support, but that’s a story from another time) since I first began using the platform four or so years ago, so I was baffled when several of them questioned its reliability. As it turns out, even now (as opposed to way back when EBS would drop out every month or two), they have good reason.

Because it’s the default, many new users often spin up their stuff in Amazon’s Northern Virginia region. However, if you look at Amazon’s own outage data, and keep in mind that us-west-2 (Oregon) has a near-identical feature set, and identical pricing, you’ll be left wondering what you’re doing in N. Virginia.

Amazon’s status page can be found here. A quick look at how she works will lead you to find data.json which contains all the outage information for the last year. After a bit of processing and some gnuplot magic, you can have a rough plot of the various outages. This is how things looked at time of writing.

data-current

While this doesn’t take into account the severity or duration of any given outage, I think it’s pretty fair to say that us-east-1 seems to be the least reliable region in AWS (even if you exclude the cluster from Sept 20th when the entire region’s API went bananas). I was lucky enough to start out using AWS with a company that was already aware of this, but for those who weren’t, consider sticking any new stuff in Oregon!

The data.json file I used for the graph above can be found here.