The document discusses the design of Percolator, a system for incrementally processing updates to large datasets. Percolator uses distributed transactions over a random-access Bigtable repository and an observer framework to incrementally update data. Transactions provide ACID semantics using timestamps from a centralized oracle. Observers register functions to be called on data mutations to trigger computation. Percolator was used to reduce the average age of documents in Google search results by 50%.