What happened after the crash, during the extended registration period?
(Source: https://www.gov.uk/register-to-vote)
For our report on the initial crash, please click here.
In brief, the web site that voters use to register crashed, after a live televised debate drove up demand to 50 times the normal level.
In this blog post we will examine what happened after the initial crash.
The political response
The immediate cross party response was to arrange an extension of the period allowed for registration. Even this was not without controversy, as various groups claimed this would unfairly disadvantage them, or open up the process to fraud.
Most politicians were focused on what happens next, though Sir Peter Bottomley observed that "The voter registration website crash was entirely avoidable."
The technical response
Computer weekly think they have the answer to why the site went down, application issues tied to poor scalability. And IT Pro are in agreement, blaming “Failures to design the voter registration website for the cloud, or to make it scalable.”
To us, these sound like credible explanations, which fit well with what we observed. The application is not doing anything very complex, (no payment processing, or complex databases), it was merely collecting information and forwarding it on.
How did the site handle the new end of registration?
The numbers behind the website performance are publicly available, and these show that demand for services during the second end period were MUCH lower. About a quarter the level at which it crashed the day before.
So it is not surprising that response times help up well. Here’s how response times fared:
What should we learn from this?
We at Capacitas think there are several key takeaways from this incident, and they are widely applicable to any business proving an online service.
Even simple services may not scale well.
While it is obviously sensible to focus on the more complex areas of your services, the simple ones should also be assessed and tested. Load testing in the live environment would have clearly revealed the limitations, and shown that the site would not fail elegantly under load.
Click here for details of service assurance services from Capacitas.
Click here for details of risk modelling services from Capacitas.
Don’t just assume the cloud will work.
Related to our first point. There is a myth that you don’t need to do capacity planning if your service is in the cloud. While the cloud CAN provide excellent and flexible service hosting, it does not guarantee service. Capacitas have an extensive track record of helping our clients handle extreme peaks, such as Black Friday. In many cases this did not even require more hardware.
Click here to read how Capacitas helped a major airline through its extreme January peak.
Make a ‘worst reasonable case’ assessment of possible peak demand.
Oliver Letwin told parliament that “The spike in registration was three times higher than for a general election.” Given that this referendum has been described by all sides as a “once in a generation decision”, the size of this peak does not surprise me.
Indeed, if you allow for the factor of a televised debate immediately before the deadline, I suspect demand was even higher than shown by those who successfully got through.
If you are not confident of the accuracy of your predictions, put defensive measures in place to manage the demand that you cannot handle.
Click here for details of Demand and Supply modelling services from Capacitas.
If you would like to learn more about our Prepare for Peak and Performance testing solutions, please click below, to see our latest Ebook.
