Factorizing a 28 billion elements matrix with Apache Flink

Use case is computing recommendations at large scale (collaborative filtering, latent factor models)