facebook stats from Dare
Dare Obasanjo has a nice summary of the Facebook Engineering Road Show.
They have some scary / fantastic stats:
Mike Shreopfer started his talk with a number of impressive statistics. These stats include the fact that 2 billion pieces of content are shared per week when you add up all of the status updates, comments, likes and other sharing gestures that are performed on the site. There are also 2 billion photos uploaded per month, 15,000 Facebook Connect implementations which in combination with the various platform applications make over 5 billion API calls per day and about 300 million active users per month.
…
They use PHP on the front-end because it is easy to learn, easy to develop in and it is very productive. However there are downsides such as the fact that their benchmarks have shown that it is ten times slower than if they used more traditional web programming technologies like C# or Java. They have done a lot of work to optimize PHP and will continue to invest in optimizations instead of switching the language they use on their front ends.
This gels with what I’ve seen at my previous employer. It’s easy to find “good enough” PHP people who also know HTML, CSS and Javascript (and not too hard to find rock stars like Lorenzo). Training someone who can code in PHP is easy, even if they feel like their brain is being extracted thru their tear ducts. That makes a lot of sense – it’s cheaper than buying more hardware by a long shot.
…
Once they got memcached throughput that high, they hit another scaling bottleneck in that they started flooding their network switches due to putting too much data on the network.
All up, it shows some of the problems that you run into when you have a site which is a lot more than just “big” – it’s “huge”. I’d LOVE to see a similar thing from Google or Microsoft. The BBC – which is, in almost anyones terms, a “big” site – doesn’t run into these kinds of problems. Google, Facebook, Microsoft, Amazon have really created a new class of scale.
