Europeana, the pan-European digital library that went down shortly after their launch, has provided some additional details about why the site was taken offline. There are some interesting numbers in the documents that were published. Let’s start with the facts that were already published:
- The site received an unexpected amount of visitors after the launch
- There were attempts to add additional capacity, but that didn’t help
- Eventually, the site was taken offline
Here are some of the extra details that were published. The number of visits sounds extremely high:
Did the traffic increase on Thursday afternoon?
The traffic increased in the afternoon and reached 13 million hits per hour (4,000 concurrent users).
All other files are “static” content; even a single server with very modest hardware can easily keep up with the 4000 hits/second that Europeana experienced. That leaves an estimated 300 to 400 “dynamic” pageviews that need to be served each second. For these, Europeana had three servers, and that capacity was increased later. Note the incorrect assumption that each hit represents a unique user:
The 3 servers employed to support Europeana in The Netherlands could not cope with the traffic of around 10 million hits per hour (3,000 concurrent users – users doing the same thing at the exact same time), which led to the slowing down or the temporary unavailability of Europeana on Thursday, 20 November.
After the site went down for the first time at about 11.00 hrs on Thursday, 20 November, the Europeana management in The Hague managed to increase computer capacity to deal with 8 million hits per hour
OK. Sounds plausible: the original capacity was 5 million hits/hour. They said capacity had been doubled, so the platform scaled pretty well. That means this would have been a beautiful example of a situation that might have benefited from cloud computing; I’m normally not a big fan of this, but if you’re launching a new site that is expected to draw 5 million visitors per hour, having an AMI image ready to start a couple EC2 instances when load on your own servers begins to get too high might be a good idea.