There is a PDF document on Google Labs that describes the model the applications of Google are implemented with. The thing is called MapReduce and is a way to code and think in algorithms that make a partitioning to a server cluster quite easy. You can do simple things like sort and grep, but after all, the large scale web index and the Google File System (GFS) are written with that model. Hey, dudes at Google, make that thing open-source. Or, why shouldn't we try to implement that thing? Anyone willing to join me?