How can large datasets be accessed in a way that lends itself to efficient processing of data in batch mode?