The original Google algorithm was based on a paper that they wrote, page rank. As the internet grew, it was harder and harder to scale Google. There was some critical point where this big batch process to index the web started failing. It would take three weeks to run the batch process. And so obviously they were freaking out inside of Google. This was the genesis for them to create MapReduce - Hadoop today.
Dalton Caldwell and Michael Seibel on software hacks that don't scale. Companies discussed include Google, Facebook, Twitch, and imeem. Watch the first video on doing things that don't scale here: https://youtu.be/4RMjQal_c4U
Apply to Y Combinator: https://www.ycombinator.com/apply/