Last year PretaWeb were asked to architect and build a new www.ses.vic.gov.au website that was designed to survive an earthquake. Not only would it need to be geographically redundant to ensure 99.99% availability but it would have to scale to handle the huge jump in traffic. You don't think such event would happen so soon after delivering on that promise, and you don't think that when it does it would be the largest earthquake Victoria has seen in a century. Two nights ago, just before 9pm AEST that's exactly what happened. Thankfully no one was hurt as the earthquake but it turned out to be the perfect validation of the site we architected.
So what does an earthquake look like to an emergency website?
![Pastedgraphic]()
A little like the quake itself, the traffic goes from nothing to massive movement instantly. In this case from nothing to 55Mbps and almost 700 requests per second. To put it in perspective, it is like a moving a website into the 1000 most popular websites list for an hour. This was done with zero downtime and no slow down in page responsiveness. In this same time period it was widely reported the Geo Sciences of Australia website (http://www.ga.gov.au) melted down.
This might seem like a mean feat but in fact the site was tested to the level of having ebay.com's traffic thrown at it instantly, 10 billion page views/month. A scenario based on an earthquake resulting in 850,000 people viewing the website within 15 minutes. It turns out that the reality was more like 10,000 visitors but it's better to be safe than sorry. To our knowledge no other emergency site in Australia has been tested to this extent.
An earthquake provides the perfect worst case scenario for scaling a website similar to the infamous "slashdot effect". Everyone gets the same signal at the same time of an event via the ground shaking. This combined with ubiquity of smart phones means within minutes a large number of people have jumped online and searched for information on what happened and what to do. It turns out that the traffic curves that occurred in reality were very similar to what was modelled in the testing, including that close to half the visits were from mobile devices.
This was the result of some very careful planning, good hosting and our technology choice, Plone, making it easy to mange some carefully crafted caching rules for all the content on the site. Plone's multi-site capabilities meant many sites on the same platform are able to take advantage of this scalable architecture.