According to one aspect, it is appreciated that it may be useful and particularly advantageous to provide a data generator that creates more realistic data for testing purposes, especially in data systems where large volumes of data are necessary. In one implementation, a data generator is provided that produces relationally consistent data for testing purposes. For instance, a synthetic data generation process may be performed that produces any number of relationally consistent data table structures. Further, in another implementation, generation of the data can be statistically influenced so that the data generated can take on the “look and feel” of production data. Also, data may be produced as needed, and its generation may be performed in parallel, depending on interdependencies in the data.