For 'true' randomness in computer programming you can use current time in fractions of a second. But yeah, for big data sets it might be more difficult.