Why didn't they just run Hadoop on a sample of the logs instead of the full dataset? It seems like they were just looking for popular related queries anyways. A

Why didn't they just run Hadoop on a sample of the logs instead of the full dataset? It seems like they were just looking for popular related queries anyways. Any ideas?