The updated cloud service availability results for Q2 2014 are in! Last quarter, we began scanning the top million websites hosted on cloud infrastructure and reported our results for Q1 2014. For a more precise account of service availability in the public cloud this quarter, we increased the monitoring resolution to an average interval of 5.63 minutes, and sampled the top million websites listed on Quantcast, scanning over 62,000 sites running on public cloud infrastructure – including Amazon Web Services, Softlayer, Rackspace, and Microsoft Azure.
Cloud Service Availability Methodology
Starting on January 1st and through Q2 2014, we monitored availability for every one of the Quantcast top 1M sites running in the cloud. Erring on the side of caution, we queried each site with an “http GET” request. If the web server returned anything but a “200 OK” result, we sent two more requests at 30 second intervals. A downtime event was registered if and only if the site’s web server did not respond with a “200 OK” result three times in a row. While it is conceivable that a web server’s availability fluctuated over 90 seconds and happened to go down for those three attempts, the more probable alternative is that every registered downtime event lasted for a duration of 90 seconds at best and 5.63 minutes at worst.
Even with the improved time interval, this methodology still paints a more optimistic picture of cloud service availability. Intervals of 5.63 minutes wouldn’t necessarily register every instance of downtime. Moreover, less than three negative results wouldn’t register a downtime event even if one did occur. On the other hand, it’s safe to assume that every registered event represents actual downtime
Who’s Up and Who’s Down in the Cloud
Across all cloud providers powering Quantcast top 1M sites, 30.1% did not record a downtime event. 9% recorded less than 53 minutes annually (13 minutes over the quarter) corresponding to 99.99% availability, 24.46% less than 263 minutes annually (66 over the quarter) for 99.95% availability, 13% recorded less than 8.76 hours annually (2.19 hours over the quarter) for 99.9% availability , and 22% recorded more than 8.76 hours annually for 99% availability or worse.
Low Cloud Presence for Microsoft Azure
While AWS, Softlayer, and Rackspace show similar availability distribution, Microsoft Azure reports significantly lower results at both extremes – only 87.8% of Microsoft Azure sites reported downtime vs. 69.9% across all clouds on average. Of course, a lower percentage of the provider’s sites reported very high downtime – 14.38% vs. an average of 22%. As a result, 54% of Azure sites reported 99.95% availability. While Azure’s dramatically low presence of less than 1% of cloud-hosted sites offers a partial sampling at best, such a dramatic difference is worth noting and warrants further testing when it comes to mission critical applications.