I talked about having six plausible options for managing the load and about the decision to choose a combination of Varnish and an additional overflow server to cope with the day’s traffic.
Setting these up was relatively painless but did involve a last minute need to migrate the whole site to a new primary server, because Varnish, for want of a better explanation, didn’t like the old one. Once that was done we were basically all set, we just had to make sure it excluded the dynamic tools we needed to run in realtime.
What was load like on the day?
Traffic levels were pretty much in line with what we expected, an increase on last year and way more than our standard infrastructure could have handled – in the end just under 50,000 sessions in 24 hours. Social media was a big driver to the site, with just under 60,000 tweets on the hashtag throughout the day – trending until about 5pm, and then again in the evening.
How did Varnish perform?
I can’t recommend Varnish enough – we set it up to manage load balancing and it handled just over two million requests through the day and barely broke a sweat, we used only 10% of our primary server’s memory, going nowhere near the overflow at all, and page load speed was a satisfying 3.2 seconds – it was perfect.
How was everything else?
Ironically after all that, our database ran into a bit of trouble around lunchtime, processing so many form requests that it crashed for around 10 minutes. But that was minor and we’ll upgrade it and increase capacity for next year – watch this space and, in the meantime, get installing Varnish!